Home

Understanding Society - UKHLS: Wave 1-2, 2009

1. 18 GENERAL POPULATION SAMPLE COMPONENT 19 GENERAL POPULATION COMPARISON SAMPLE 19 ETHNIC MINORITY BOOST SAMPLE ctv en den RR die ER EHE EUR 20 FORMER BHPS SAMPLE a an aun a C RR X a D 21 SAMPLE STATUS AND FOLLOWING RULES essen nnne nnne 21 SAMPLE DESIGN VARIABLES AND ANALYSIS seeeeee eene nnne nnn 22 WEIGHTING ADJUSTMENTS FOR THE WAVE 2 23 SELECTING THE CORRECT WEIGHT FOR YOUR ANALYSIS eee 24 NAMING CONVENTIONS FOR WEIGHTING VARIABLES 2 0cccccccceeeeeeeceeeeeeeeeeeeeeees 27 TEGHNICAUDETAILS oia EP E E 28 IMPUTATION OF INCOME VARIABLES 2 ein 36 WHAT DO WE METTE Tiii tei ds ter ce ties bri am ate QC 37 IMPUTATION PROCEDURES ss0scccsssessccesssssecessnsnsecesecsnsecesusaasanecssssansecssssansecesees 38 ITEM NON RESPONSE ON INCOME VARIABLES IN THE INDIVIDUAL QUESTIONNAIRE 38 ITEM NON RESPONSE FOR INCOME VARIABLES IN THE PROXY QUESTIONNAIRE 40 INDIVIDUAL NON RESPONDENTS WITH NO PROXY 40 CODING sec CT ET 41 3 FILE AND VARIABLE INFORMATION cccecceeceeeeececseeeececeeeeeeueeeueeeneeuneeans 41 INFORMATION ABOUT THE BHPS SAMPLE COMPONENT ccccccececeeeeceeceeeeeeeeeeeeeeeeeeeeess 41 VARIABLE INFORMATION OVERVIEW BASIC A
2. W HHSAMP This includes data on the area surrounding the address the type of accommodation and other information that the interviewer can observe about sampled addresses Reasons for refusal are also available Interviewers also collect some information 51 about the quality of the interview and persons present during the interview process This is available along with substantive data collected during adult individual interviews including proxy interviews in W_INDRESP 4 Data Access We release the preceding waves of data when we make a new edition available There have been some corrections to the earlier wave The user should refer to the document made available by the UK Data Service SN 6614 Understanding Society Wave 1 2009 2010 Revision 1 2012 for details We request that researchers using the data notify us about errors inconsistencies and other problems with the data identified during their use of the data We make use of this information in improving the data Please raise any issues relating to data or data analysis with our user support service http data understandingsociety org uk documentation support We will communicate information to members of the Understanding Society users group or via Frequently Asked Questions on the Understanding Society web page about data http data understandingsociety org uk The data are released through the UK Data Archive UKDA in SPSS and Stata formats While documentati
3. Contents TINTRODUCTION ee tl OVERVIEW OF STUDY FA nt EXE sas FARK QNA KUNA tes ROUTE GUIDE FOR USERS ocu ua casa dan 4 Ed ua is 4 DATA COLLECTION AND RESPONSE OUTCOMES eene nennen nennen nennt nsns nn nnn nnns 5 OVER VIEN AS Ata Mt orto en ta Mad eta 5 FIGURE 1 TIMING OF DATA COLLECTION 5 DATA COLLECTION E KRUGER ESO REDE ECL ESTE Cd 5 PANEL MEMBERSHIP AND PANEL MAINTENANCE eeeeeeennnn eene nnne nnne nnn 7 RESPONSE OUTCOMES WAVE scettur tente d te nbn ch cat ome bns 8 RESPONSE OUTCOMES WAVE 2 i ira tek re Re dte en Fl etr rx Rt oe tu aee uA rud tg dd 8 DATA PROCESSING AND CEEANING ies eee ed legs Rr ee ped re Eve 13 DOCUMENTATION OF THE QUESTIONNAIRES MODULES AND 13 READING THE QUESTIONNAIRES 15 eet back oa vt t n Eco aM 14 SUMMARY OF QUESTIONNAIRE MODULES sienne eene nnne nennt nnn 15 CHANGES TO THE QUESTIONNAIRE eese eese nnne asas aaa ansa saa ssa 18 OTHER FIELDWORK MATERIALS eene seen rnn nnn hn nennt asses tans satanas san 18 AMPLE DESIGN f T
4. The imputation by chained equations ICE starts by considering the following recursive triangular system of imputation equations where are the income and auxiliary variables to be imputed ordered from the one with the smallest percentage of missing values Y4 to the one with the largest percentage of missing values Y X is a set of auxiliary variables observed for all individuals a s and are parameters and Uj Up Ux are random errors Such a recursive system allows us to carry out the imputation separately for each variable and sequentially The sequential procedure is given by the following steps 1 estimation of the first equation and imputation of the missing values for Y4 2 estimation of the second equation using the imputed values to replace the missing values of Y and imputation of Y2 3 repetition of estimation and imputation steps sequentially for each of the following equations until when all k variables Y Ys Y have been imputed We use stochastic imputation that is we draw the imputed values from the posterior predictive distribution of the variable to be imputed conditional to the observed data For more details about stochastic imputation we refer to Rubin 1987 Schafer 1997 and Kenward and Carpenter 2007 39 This sequential estimation is consistent only if the recursive system is valid Since this is not necessarily a valid assumption ICE uses the imputed values produc
5. In the following we describe the imputation by chained equations ICE adopted for item non response in the individual personal and proxy questionnaires and for individual non response that is for those for whom there is neither an individual nor a proxy questionnaire available Item non response on income variables in the individual questionnaire The imputation of income variables in the individual questionnaire is performed considering a separate equation for each of the income components including each of the sources reported in the income data file We use log linear models for each of our income variables The explanatory variables are a set of characteristics collected in the individual personal or household questionnaires The specification of the models varies by income variable but it generally includes the following variables e personal socio economic variables age sex self reported ethnic group indicator for respondent born in the UK marital status education level general health current subjective financial situation personal income variables excluding the one used as the dependent variable e household characteristics number of children in the household house tenure house type household size e job characteristics log number of hours normally worked per week log number of hours per months in a second job log years of job tenure permanent or temporary job occupation soc 2000 1 digit number 3
6. participants understandingsociety org uk Response outcomes Wave 1 The Wave One mainstage fieldwork started on 8 January 2009 and ended on the 7 March 2011 including the re issue period In total interviews were achieved in 30 169 households 26 089 in the General Population Sample 4 080 in the ethnic minority boost sample with full or proxy interviews with 50 994 individuals 43 674 in the General Population Sample and 7 320 in the ethnic minority boost sample Tables 1 and 2 below present the household and individual response rates for Wave 1 The individual response rates are for co operating households only Table 1 Household response rates among eligible households Wave 1 General Population Sample ee inorit Great Britain Total E Responding 57 196 60 9 57 3 39 9 Non contact 8 1 11 0 8 3 28 0 Refusal 33 9 27 496 33 6 29 0 Other 0 8 0 7 0 8 3 1 N 43 267 2 107 45 374 10 077 The response rates for the ethnicity boost sample component do not make any correction for the probability of non interviewed cases being ineligible The estimated response rate taking this factor into account is substantially higher Table 2 Individual response rates Wave 1 General Population Sample Ethnic Great Britain pel Total Minority Boost Full interview 82 0 77 3 81 8 72 4 Proxy interview 5 3 3 5 5 2 6 9 Refusal 6 5 9 2 6 7 8 7 Other non interview 6 1 9 9 6 3 12 1 N 47 615 2 584 50 1
7. to which the sample member belongs The prefix w denotes waves in general The value of w psu does not change between waves but for new sample entrants it is only defined from the wave at which they enter the sample w psu takes values in the following ranges 1 575 former BHPS sample in Identical to the BHPS variable wpsu England Scotland and Wales 701 1849 former BHPS Northern Corresponds to initial BHPS wave 11 sampled Ireland sample households as these were selected in a one stage design 2001 4640 UKHLS GPS in England Corresponds to the postal sectors used as Scotland and Wales PSUs see Lynn 2009 4644 7035 UKHLS GPS in Northern Corresponds to wave 1 sampled households Ireland as these were selected in a one stage design 7078 51776 UKHLS EMB Corresponds to wave 1 sampled households as these were selected in a one stage design within the high minority density domain see Berthoud et al 2009 22 w_ strata This indicates the sampling stratum from which the sample member was selected The value of w_strata does not change between waves but for new sample entrants it is only defined from the wave at which they enter the sample w_strata takes values in the following ranges or the value 701 for the ex BHPS Northern Ireland sample 1 151 ex BHPS sample in Identical to the BHPS variable wstrata England Scotland and Wales 701 ex BHPS Northern Northern Ireland treated as a single stratum Ireland
8. we have drawn upon the documentation for the British Household Panel Survey Taylor 2010 see also http www iser essex ac uk bhps 2 Study Related Information Design overview Understanding Society is a panel survey of households with yearly interviews Data collection for a single wave is scheduled across 24 months The study began with a representative probability sample of households There is an extended discussion of sample design below and in Lynn 2009 Adult household members age 16 or older are asked questions and the same individuals are re interviewed in successive years to see how things have changed Household members aged 10 15 years are asked to complete a short self completion youth questionnaire Children become eligible for a full interview once they reach the age of 16 The overall study has multiple sample components In the mainstage survey there is the a General Population Sample with its subset the General Population Comparison Sample b the Ethnic Minority Boost Sample and c participants from the British Household Panel Study The instruments for the first three components are the same except the EMB sample and the General Population Comparison sample have an Extra five minutes of questions specifically relevant to ethnic minority communities e g ethnic identity and remittances In addition there is a separate survey the Innovation Panel IP which is fielded in the year before the mainstage
9. Others were computed for the purpose of analysts Analysts should consult the description of derived variables that they plan to use in their analyses The derived variables are documented on the detailed variable view on the Understanding Society website The documentation summarises the variables used in the computation of the derived variable See the detailed view for a scghq2 dv a categorical or caseness expression of scores for the GHQ 12 as an example Proactive dependent interviewing was used in Wave 2 and will continue to be used in subsequent waves to increase efficiency of data collection and lessen respondent burden Specifically information reported at an earlier time is fed forward to the respondent to personalize the question So rather than ask a question about current occupation with its complex probing by interviewers the question might say the last time you were interviewed you said you were specific occupation are you still specific occupation Feed forward variables are used at both the household and individual levels For example b ff hhsize feeds forward the household size from the previous wave Wave 1 The variable b ff plbornc is the country of birth of the respondent fed forward from the previous wave 45 Note the use of the prefix ff Some of the fed forward variables were not used in the wording of a question but were used by the CAPI script to route respondents appropriately based on information fr
10. UK The household response rate for the continuing Understanding Society General Population Sample was 76 8 in Great Britain and 81 9 in Northern Ireland The household response rates for the former BHPS samples were similar to the Understanding Society General Population Sample Among the samples in Great Britain the Living in Britain households had the highest response rate at 77 2 The Living in Wales households had a similar response rate to the Living in Britain sample 76 8 whilst Living in Scotland had a lower response rate at 73 5 The NIHPS had a higher household response rate than the Great Britain samples with 84 8 The response rates for the BHPS samples in Great Britain were disappointing given that this was in effect Wave 19 for many households However the lower response rate may have been due to the change in the fieldwork agency in interviewers in the survey name and in the logo Interestingly in Northern Ireland where the survey name and logo changed but the fieldwork agency and so the interviewer stayed the same as in NIHPS the response rate was much higher Table 3 Household response rates Wave 2 UKHLS GP EMB Former BHPS sample Living Living UKHLS UKHLS Living in in in NIHPS Total GB NI Britain Scotland Wales Fully 16 003 873 2 030 3 112 793 833 990 24 634 responding 61 8 65 396 49 396 66 5 64 3 64 1 73 4 61 7 Partially 3 888 221 749
11. earnings and total gross income We impute missing values for these two variables again using ICE The imputation is based on the sample of persons responding to the individual questionnaire where missing values have been replaced with the imputed values produced by ICE as explained in last section together with the sample of individuals for whom a proxy questionnaire is available The imputation process is comparable to the one described in last section Since individuals answering the proxy questionnaires are asked to report income brackets rather than point values we use interval regressions for both earning and income We impute total gross earnings and total gross income using the explanatory variables described above Individual non respondents with no proxy questionnaire For individual non respondents with no proxy questionnaire but in responding households we use information from the household questionnaire to impute a total personal income The procedure used is again the imputation by chained equations ICE We first impute the total gross income then we impute the total net income using gross income as a predictor in addition to the other explanatory variables The user should notice that the imputation of personal income for individuals for whom there is neither a personal nor a proxy questionnaire is based only on variables available in the household questionnaire More precisely we use e individual socio economic variables age s
12. groups identified by the screen in order to select only a desired proportion into the sample non mixed Indian African Far Eastern Middle Eastern For other target groups all resident persons were included in the sample mixed Indian Bangladeshi mixed Caribbean Sri Lankan Chinese Turkish e In households included in the sample in the previous two steps all members of target ethnic groups were deemed to be members of the Ethnic Minority Boost sample including children All persons of other ethnic groups are not Ethnic Minority Boost 20 sample members They will be interviewed as temporary sample members for so long as they remain co resident with at least one Ethnic Minority Boost sample member The overall sampling fractions combine a the probability of sampling the sector b the fraction of addresses selected within the sector and c the probability of a household being retained following the application of the random selection mechanism described above Former BHPS sample The sample issued at Wave two consisted of all members from the BHPS sample who were still active at Wave 18 of the BHPS and who had not refused consent to be issued as part of the Understanding Society sample It should be noted that the BHPS sample contains different components including the original sample where households were first selected in1991 boost samples in Scotland and Wales first selected in 1999 and a Northern Ireland sample selected
13. in Wales and the remaining 758 were in England with a concentration in London 412 sectors The number of addresses selected per postal sector ranged from 15 to 103 Sampling fractions varied across the sectors in a way designed to deliver target numbers of respondents in each target ethnic minority group with adequate statistical efficiency see Berthoud et al 2009 for more details In sectors selected for both the General Population Sample component and the Ethnic Minority Boost sample a single systematic sample of the required total number of addresses was selected and allocated in a systematic way to the two sample components thus ensuring that both sample components are spread throughout the whole sector The final stage of sampling was done by the interviewers for the Ethnic Minority Boost sample though its procedures were somewhat more complex You can see the steps described in the Project Instructions for Interviewers http data understandingsociety org uk assets 476 At addresses containing more than three dwellings or households the procedures to sub select dwellings or households were as described above for the General Population Sample component Within each household rather than all resident persons becoming sample members there were three additional steps e A screen was carried out to identify whether there were any persons from target ethnic groups in the household e Arandom mechanism was applied to certain target
14. observed with a single parent in a household in the first wave after their birth the weight given was the parent s weight This reflects a close to zero likelinood for the baby to be sampled via the other parent The adjustment for household non response at UKHLS Wave 2 was derived from a model of enumeration at Wave 2 conditional on entering the UKHLS sample i e being issued to the field for UKHLS Wave 2 in which covariates came from the Wave 9 household instruments for England Scotland and Wales and the Wave 11 household instruments for Northern Ireland The weight which reflects the chance of a BHPS OSM of being selected into the BHPS to be issued into UKHLS and to be enumerated at Wave 2 of UKHLS is the BHPS 2010 longitudinal enumerated person weight D psnenbh lw Finally the BHPS cross sectional enumeration weight b_psnenbh_xw was created through a weight share method by sharing the BHPS 2010 longitudinal enumerated person weight to TSMs and PSMs The BHPS cross sectional weights for main proxy or telephone interview respondents b_indpxbh_xw main interview respondents b_indinbh_xw and self completion respondents adults b indscbh xw and youth b ythscbh xw each consist of the cross sectional individual enumerated weight with an additional adjustment for non response to the relevant instrument conditional on household response These adjustments were based on logistic regression models with both individual level and house
15. participants in successive waves as long as they live in the household of an OSM A male TSM who fathers a child with an OSM female becomes a Permanent Sample Member PSM PSMs are treated in the same way as OSMs in the following rules In sum TSMs are not followed for interviews when they leave the household but OSMs and PSMs are An exception to these sample status rules is that at Wave 1 individuals who were not of an ethnic minority within an ethnic minority boost household were classified as TSMs For panel maintenance ISER maintains a database of information on respondents so we can send communications to them and to allocate interviewers This information is vital for minimising attrition The data base builds on contact information collected during the survey interviews and is updated throughout the year There are for example new addresses household splits and moves out of the country or into an institution Change of address cards were also returned to ISER in cases where a whole household moved or a new resident returned the card giving the forwarding address Finally it is possible for ISER to be notified of some deaths through this means A between wave mailing is also used to help maintain contact with participants and update addresses The mailing has a report of research findings an address confirmation slip and materials to encourage registration with the Participant website The participant website can be seen at http
16. sample 2001 3320 UKHLS GPS in England Corresponds to groups of two or more PSUs in Scotland and Wales selection order as they were selected systematically from an implicitly ordered list see Lynn 2009 3321 UKHLS GPS in Northern Northern Ireland treated as a single stratum Ireland 3322 5117 UKHLS EMB Corresponds to the postal sectors in the high minority density domain as selections were made independently from each see Berthoud et al 2009 Example using Stata In Stata to obtain estimates that correctly take into account the sample design the user need only specify the design variables using the svyset command for example svyset a psu pweight a indpxus strata a strata Then any compatible command simply needs to be prefixed with svy for example svy logistic depvar variable1 variable2 variable3 Weighting adjustments for the Wave 2 release A number of weights are provided for data users They adjust for unequal selection probabilities differential nonresponse and potential sampling error A weighted analysis will adjust for the higher sampling fraction in Northern Ireland and for different probabilities of selection in the Ethnic Minority Boost sample as well as for response rate differences between subgroups of the sample Separate sets of weights are provided for the GPS and EMB sample component and the ex BHPS sample Considering the complexity of the study design weights should be selected carefully follow
17. this question And If LChLv 7 Child resident And If resp is biological mother of resident child Resp is biological mother of resident child And If resp is biological mother of resident child amp child lt 16 Resp is biological mother of resident child under 16 Summary of questionnaire modules About half of the questionnaire content is collected annually with additional modules collected different intervals often every two to three years The long term content plan summarizes the pattern that has been collected or planned http www understandingsociety org uk design content outlines aspx Table 6 Summary of Questionnaire Modules in Waves 2 and 1 Module Asked of a subset Wave 2 Also in Wave 1 in Wave 2 Demographics new entrants X X of all Initial Conditions new entrants X X of all Own First Job new entrants X X part of excluding rising 16 s employment status history asked in months 1 6 Parental Education if not asked in Wave X X 1 Educational asked of full time X Aspirations students Young adults age 16 21 X Family background new entrants X Ethnicity and new entrant X X of all National Identity Childhood Language X Ethnic Identity EMB GPC LDA X 15 Religion new entrant and X X of all EMB GPC LDA General Health X X see Health and Disability module N
18. variable is imputed In most cases this takes the value 1 if imputed and 0 if not but in the case of the following variables it shows the proportion of total income imputed fimngrs if fibenothr if and w fihhmngrs if In the income data file there may be multiple receipts of income from the same source For example a respondent may have multiple pensions from a previous employer These are imputed in a single step together and the imputed values are in the variable w frmnthimp This variable is set to inapplicable for the second and subsequent receipt of income from a single source Imputation procedures The procedure used in Understanding Society is imputation by chained equations ICE Each income variable is imputed by stochastic regression imputation using as predictors a large set of auxiliary variables which includes income variables and other potential correlates such as personal and household socio demographic characteristics Some of these characteristics are missing and must also be imputed but the released data contains imputed values only for the income variables Imputation by chained equations ICE allows for interdependence between income and auxiliary variables by considering univariate models estimated separately and sequentially see Van Buuren et al 1999 and Ragunathan et al 2001 This method has been already used in some major household panel surveys such as the European Community Household Panel Survey
19. was implemented as described above in the individual level enumeration weight section except that a greatly reduced matrix was used in the case of the Extra five minutes weight due to the much smaller sample size for which this weight applies After multiplying by the poststratification adjustment each of the obtained weights was then scaled to a mean of one UKHLS longitudinal weights Each of the five types of longitudinal weights enumerated persons proxy or main interview main interview self completion and Extra five minutes interview is based on the corresponding Wave 1 cross sectional weight with an additional adjustment for non response at Wave 2 Each adjustment was based on a model of Wave 2 response conditional on Wave 1 response to the relevant instrument For the enumerated persons 32 model covariates were taken from the Wave 1 household grid and household questionnaire In the model for proxy and main interviews covariates were taken from the Wave 1 proxy interview or the equivalent items from the main interview household grid and household questionnaire In both the model for main interviews and the model for adult self completion questionnaires covariates were taken from the Wave 1 main interview household grid and household questionnaire The adjustment weight was calculated as the reciprocal of the model predicted response propensity The Wave 1 weight had already adjusted for differential selection probabili
20. were then allocated systematically to 24 monthly samples with 110 sectors in each monthly sample Within each postal sector 18 addresses were selected using systematic random sampling The England Scotland and Wales sample in this data release is therefore based upon an initial sample of 47 520 addresses In Northern Ireland 2 395 addresses were selected in a single stage from the list of domestic addresses In combination this data release is therefore based upon a total of 49 915 addresses At each address the final stage of sampling was carried out by field interviewers This consisted of identifying persons to be defined as sample members All persons resident at each sample address at the time the interviewer made contact were deemed to be a sample member with the exception of the small proportion of addresses that contained more than three dwellings or households In those cases three dwellings or households were sub sampled at random General Population Comparison Sample component The General Population Comparison Sample GPCS has one sampled address for 4096 of the selected postal sectors in General Population Sample GPS component for Great Britain In other words of the 2 640 general population sectors 6096 of them 1 584 contain 18 GPS addresses and the other 40 contain 17 GPS addresses and one GPCS address The persons in these households will be designated as members of the General Population Comparison sample regardles
21. 15 11 527 0 8 1 2 1 4 0 7 0 7 0 7 0 5 0 8 Household non contact 1 890 68 500 376 126 96 34 3 092 4 7 3 4 7 4 4 6 6 4 4 3 1 5 4 9 Household refusal 4 144 167 734 965 245 260 109 6 633 10 4 8 3 10 9 11 9 12 4 11 7 4 8 10 5 Household other non interview 65 2 31 18 3 2 2 123 0 2 0 1 0 5 0 2 0 2 0 1 0 1 0 2 Household untraced 2 252 63 639 439 105 140 105 3 744 5 6 3 1 9 5 5 4 5 3 6 3 4 7 5 9 Household ineligible 507 22 208 280 51 86 68 1 222 1 396 1 196 3 196 3 5 2 6 3 9 3 0 1 9 Total 39 929 2 025 6 751 8 118 1 970 2 229 2 251 63 285 12 Data Processing and cleaning NatCen delivers the data for a sample month to ISER in batches Delivery is scheduled for 4 months following the beginning of the fieldwork process to allow time for interview re issue coding and data entry from paper documents e g the self completion instruments Data is delivered as SPSS system files which are then exported to triple S data exchange format and imported into a SIR database Quality control processes include extensive data checking to ensure that the data conform to the expected structure and to the routing and range constraints defined by the questionnaire specifications Data anomalies are investigated to determine whether they are related to 1 the invalid specification of the questionnaire 2 the incorrect scripting of the questionnaire 3 a failure to specify that a particular constraint shoul
22. 504 114 165 153 5 794 responding 15 0 16 5 18 2 10 8 9 2 12 7 11 4 14 5 All 19 891 1 094 2 779 3 616 907 998 1 143 30 428 responding 76 8 81 9 67 5 77 2 73 5 76 8 84 8 76 2 Non contact 1 116 22 299 217 73 62 33 1 825 4 3 1 7 7 3 4 6 5 9 4 8 2 5 4 6 Untraced 1 450 50 411 181 49 50 43 2 235 mover 5 6 3 7 10 0 3 9 4 0 3 9 3 2 5 6 Refusal 3 359 162 600 648 199 185 117 5 281 13 0 12 1 14 6 13 8 16 1 14 2 8 7 13 2 Other non 94 8 28 20 6 5 12 173 interview 0 4 0 6 0 7 0 4 0 5 0 4 0 9 0 4 Total 25 910 1 336 4 117 4 682 1 234 1 300 1 348 39 942 Base is all households issued to the field for wave 2 minus any found to have become ineligible Non contact rates were lower in Northern Ireland than in Great Britain The level of untraced movers was higher for the Understanding Society General Population Sample in Great Britain than in the former BHPS samples The level of non contact and untraced movers were highest in the ethnic minority boost samples possibly reflecting the younger average age of this sample the concentration in large urban areas and the higher level of mobility Within the former BHPS samples the level of untraced movers was higher than in the past This is likely to be due to the increased gap between waves of interview The interviews for the former BHPS sample for Wave 2 of the UKHLS took place throughout 2010 and into the
23. 58 3 870 5 1 3 1 5 7 2 5 2 0 3 1 2 1 4 6 Telephone interview 202 66 58 326 E i 2 0 2 6 2 1 i 0 496 Other non interview 1 184 126 472 200 58 64 107 2 211 2 2 4 4 4 4 2 0 2 3 2 396 3 896 2 696 Refusal 2 104 218 511 341 92 114 133 3 513 4 096 7 7 4 8 3 4 3 7 4 1 4 8 4 1 Household non contact 3 338 125 1 156 555 210 155 111 5 650 6 3 4 4 10 8 5 6 8 3 5 6 4 0 6 7 Household refusal 7 229 350 1 743 1 493 400 427 203 11 845 13 6 12 3 16 2 15 0 15 9 15 4 7 2 14 0 Household other non interview 118 5 60 29 9 6 4 231 0 2 0 2 0 6 0 3 0 4 0 2 0 1 0 3 Household untraced 4 178 159 1 207 754 172 208 178 6 856 7 9 5 6 11 2 7 6 6 8 7 5 6 4 8 1 Total 53 254 2 840 10 742 9 967 2 517 2 769 2 802 84 891 11 Table 5 Longitudinal individual re interview rates adults by sample origin Full interview at the previous wave UKHLS GP sample EMB Former British Household Panel Survey UKHLS UKHLS Living in Living in Living in GB NI Britain Beollard Wales ES Total Full interview 29 646 1 640 4 200 5 633 1 335 1 507 1 875 45 836 74 3 81 0 62 2 69 4 67 8 67 6 83 3 72 4 Proxy interview 775 16 188 97 17 38 15 1 146 1 9 0 8 2 8 1 2 0 9 1 7 0 7 1 8 Telephone interview 184 59 57 300 i i 2 396 3 096 2 696 E 0 596 Other non interview 334 22 157 73 16 28 32 662 0 896 1 196 2 396 0 996 0 8 1 3 1 4 1 1 Refusal 316 25 94 53 13
24. 8 employed at the current job workplace for employees number of employees if self employed whether is self employed and hires employees whether the employment organization is private or not only for employees type of ownership if self employed sole ownership or partnership an indicator for whether annual business accounts are prepared for the Inland Revenue for tax purposes if self employed e household variables reflecting economic situation log amount spent on food from food shops in four weeks prior to interview log amount spent on food eaten outside the home in four weeks prior to interview log last year expenditure on domestic fuel e g electricity and gas number of bedrooms in the house number of other rooms in the house Council Tax band e government office regions At Wave 2 we use the value for the same variable at Wave 1 or Wave 18 of the BHPS where this is available Furthermore we use additional regression models to impute explanatory variables when missing More specifically we use log linear regression for continuous variables and binary ordered and multinomial logit models respectively for dummy ordinal and unordered categorical variables Finally we consider interval regression when we have brackets rather than point information or when we have a priori information which allows us to bound the missing income variable This is the case for dividends and interest for which we have bracketed information
25. 99 9 237 Response outcomes Wave 2 Table 3 below shows the household response rates for Wave 2 of the UKHLS The table separates the different samples The General Population Sample GPS consists of respondents in Great Britain and Northern Ireland The ethnic minority boost EMB households are only located in Great Britain The former BHPS sample consists of the Living in Britain sample started in 1991 the Living in Scotland and Living in Wales boost samples started in 1999 and the Northern Ireland Household Panel Survey NIHPS started in 2001 also a boost sample Ineligible households have been removed from the table these would include households where all sample members had died consist of only TSM individuals or emigrated from the UK For the former BHPS samples ineligible households would also include households which have merged with a previous wave household for example an adult moving back to live with his or her parents who are also part of the sample Responding households are those in which the household is successfully enumerated the household questionnaire is completed and all eligible adults give an individual interview Partially responding households are those where the household is enumerated and a household questionnaire is done and at least one eligible adult but not all eligible adults complete an individual interview Household response rates were higher in Northern Ireland than in the rest of the
26. Ireland e that people who live at an address with more than three dwellings or more than three households are the same as those who don t e that people who responded at Wave 2 of UKHLS in 2010 are the same with respect to your estimates as those who may have become non respondents at any time since Wave 1 of BHPS in 1991 We therefore strongly suggest conducting weighted analyses of the Understanding Society data Naming Conventions for Weighting Variables The naming conventions for the will help users to select the weight they need or to interpret the purpose of some weight variables The structure is as follows Xxxxyyzz aa where w wave xxx target population yy instrument zzz sample aa weight type options for xxx hhd household psn persons 0 ind persons 16 yth persons 10 15 options for yy en enumeration grid in interview 27 px interview or proxy 5m extra 5 minutes items sc self completion ns nurse visit bd blood options for zz us the GPS and Ethnic Minority Boost of the UKHLS sample bh BHPS sample 91 BHPS original sample starting in 1991 England Scotland and Wales 01 BHPS sample starting in 2001 original sample Scotland and Wales boost NI ip Innovation Panel Options for weight type aa Iw longitudinal analysis weight xw cross sectional analysis weight Id longitudinal design weight xd cross sectional design weight li longitudinal inclusion wei
27. ND DERIVED 43 VARIABLE NAMING AND LABELLING 43 LEARNING ABOUT THE STUDY VARIABLES 43 IDENTIFIERS AND USEFUL VARIABLES sseeeeeeee eene nennen nennt ennt n na 44 TABLE 9 SOME USEFUL VARIABLES i isset a 44 DOCUMENTATION OF DERIVED 45 EXAMPLE CODE FOR MATCHING FILES 46 EXAMPLE 1 DISTRIBUTING HOUSEHOLD LEVEL INFORMATION TO INDIVIDUAL LEVEL 46 EXAMPLE 2 SUMMARISING INDIVIDUAL LEVEL INFORMATION AT THE HOUSEHOLD DAWN M M 47 EXAMPLE 3 MATCHING INDIVIDUALS WITHIN A HOUSEHOLD eene 48 EXAMPLE 4 USING THE EGOALT FILE TO CREATE HOUSEHOLD COMPOSITION VARIABLES 49 EXAMPLE 5 MERGING INDIVIDUAL FILES ACROSS WAVES INTO LONG FORMAT 49 EXAMPLE 6 MERGING INDIVIDUAL FILES ACROSS WAVES INTO WIDE FORMAT 50 PRESERVING CONFIDENTIALITY ceeeenen eem em ene ene nne rhse rne rre tre rest 51 x rp p MT 51 4 za sates 52 5 CITATIONS AND ACKNOWLEDGEMENTS eceenn eene nennen nenas 53 Oo BIEFERENGES oed ttr Ea
28. NS mid year estimates with a correction for institutionalized population and poststratification was implemented for the fully crossed matrix of gender by geographical region by 5 10 year age groups Thus the individual level enumerated weight consists of The individual level design weight the household nonresponse correction the poststratification adjustment The obtained weight is then scaled to have a mean of one Individual Level Nonresponse Adjustment 31 Five different individual level weights were prepared for users reflecting nonresponse occurring at different levels and different questionnaire instruments Each individual level weight consists of The individual level design weight the household nonresponse correction the individual level nonresponse correction conditional on household response the poststratification adjustment The individual nonresponse correction conditional on household nonresponse is modelled at three levels e For adult respondents age 16 or older who either completed the main interview or for whom a proxy interview was completed for a_indpxus_xw e For adult respondents age 16 or older who completed the main interview only for a indinus xw and a indbmus xw e For respondents aged 10 or older who completed and returned the self completion questionnaire for a indscus xw and a ythscus xw Note that the same model was used for respondents regardless of whether they were selected i
29. SN6849 Understanding Society Innovation Panel Waves 1 2 2008 2009 http www esds ac uk findingData snDescription asp sn 6849 The Ethnic Minority Boost sample was undertaken to produce enough cases to study households and individuals from five major ethnic groups in the UK The boost sample receives an additional five minutes of questions related to content areas that may particularly involve them The General Population Comparison sample component is also asked these questions As an introduction to the data and documentation we recommend the following reading 1 The summary of the general questionnaire content Section 2 Documentation of the questionnaires modules and questions Reading the questionnaire and notes on naming conventions Section 3 Variable naming and labeling conventions 2 Study level information is in Section 2 This includes sections on sample design weighting adjustments and data collection and response outcomes 3 Variable level descriptions of the data can be found on the Understanding Society website http data understandingsociety org uk documentation The online documentation has extensive links between questions and detailed views of variables and data files There is also a search facility for searching questions variables modules and datafiles 4 The example Stata code for matching variables from different records Section 3 Example code for matching files In assembling the documentation
30. Wales the information was linked at Middle Layer Super Output Area MSOA or Lower Layer Super Output Area LSOA levels and was obtained from http neighbourhood statistics gov uk The examples of linked information obtained from Census 2001 include the proportions in the MSOA of employed retired outright property owners travellers to work using different types of transport single household members households with one car people with different types of qualification and professional occupation among others Other linked information includes 2010 information on multiple deprivation indexes on crime instances 2009 information on inflow and net change of neighbourhood population and the proportion of different allowance claimants and 2008 information on hospital admissions and energy consumption For Scotland the information was linked at the data zone level from http www scrol gov uk scrol common home jsp and from http www scotland gov uk T opics Statistics SIMD From the Census 2001 information was obtained on population density mean age average household size and number of rooms per household in the data zone as well as the proportions in the data zone born in Scotland and outside the EU of different religious denominations employed unemployed and retired disabled those with different levels of qualification and types of occupation and different types of accommodation among others For Northern Ireland the information was lin
31. ach briefing along with two or three briefing managers or area managers The briefings were led by at least one researcher from NatCen with the majority also attended by ISER staff The briefings in Wave 1 took place across the UK Belfast Birmingham Brentwood Bristol Derby Edinburgh Glasgow Leeds London and Manchester Similar topics and and locations were used for the Wave 2 briefings These one day briefings had morning sessions devoted to fieldwork procedures including dealing with the administrative forms to record contact information and how to deal with the complexities of multiple dwelling units and multiple households The afternoon was spent discussing the survey content and reviewing and working with the Blaise computer aided personal interview CAPI instrument Interviewers were assigned to specific areas For Wave One 911 interviewers were employed to cover 3 517 areas in the sample The number of interviewers briefed in Wave 2 was 819 Fieldwork Fieldwork for Wave 1 was somewhat different because we did not know who was in the sample Before contacting any of their sample in Wave 1 interviewers mailed an introductory card from ISER to all sampled addresses addressed to The Occupier together with a small leaflet outlining the purpose of the survey The interviewer called within a week of the mailing At the end of the first interview all participating households received a more detailed brochure giving further inf
32. ariables and flagging issues in using the data They include Jakob Petersen Cara Booker Alexandra Skew Mark Bryan Mark Taylor and Alita Nandi 6 References Berthoud R Fumagalli L Lynn P amp Platt L 2009 Design of the Understanding Society ethnic minority boost sample Understanding Society Working Paper No 2009 02 Colchester ISER University of Essex http research understandingsociety org uk publications workingpaper 2009 02 Kenward M and J Carpenter 2007 Multiple imputation current perspectives Statistical Methods in Medical Research 16 3 199 218 Lynn P 2009 Sample Design for Understanding Society Understanding Society Working Paper 2009 01 Colchester University of Essex http research understandingsociety org uk publications working paper 2009 01 pdf Lynn P Burton J Kaminska O Knies G and Nandi A 2012 An initial look at non response and attrition Understanding Society Working Paper 2012 02 Colchester University of Essex http research understandingsociety org uk publications working paper 2012 02 Lynn P and O Kaminska 2010 Weighting strategy for Understanding Society Understanding Society Working Paper 2010 05 Colchester University of Essex http research understandingsociety org uk publications working paper 2010 05 53 Office of National Statistics 2010 Midyear population estimates 2009 June 24 2010 Edition http www statistics gov uk
33. ation On the study website we are presenting the questionnaire in PDF format Similar to other PDF documents the text of the questionnaire can be searched for specific words such as variable names or words in questions They are also convenient for printing sections of the instrument The self completion instruments are also displayed in pdf to correspond to the way they appeared to participants except they have been annotated with variable names The principal adult questionnaires are organized in modules Modules can be searched for in the online documentation system In the pdf formatted questionnaire clicking on entries in the table of contents will advance you to the beginning of that module 13 Instruments and survey materials were translated into multiple languages Bengali Punjabi in Urdu and Gurmukhi scripts Welsh Arabic Somali Cantonese Urdu and Gujarati Translated documents can be requested by email from info understandingsociety org uk Reading the questionnaires Figure 1 shows a marked up sample page providing information for how to interpret the questionnaire text Note that the variable names in the questionnaire do not have the wave prefix a_ Figure 1 Mark up of household questionnaire Variable name and Variable label Hsownd House owned or rented Note that there is no prefix Must add prefix to the variable name ae This variable has also been in the BHPS Text Does your household own this accomm
34. bove design weight but differs in the following ways It adjusts for the fact that the GPS Comparison Sample is only 1 45 of the GPS original sample that all EM members in low density areas were administered the Extra five minutes and that EM members in high density areas had a chance to be selected into either the GPS Comparison sample or the EMB Similar to the above weight non EM persons were assumed to have a chance to be part of only the GPS Comparison Sample and not part of the EMB Household level Nonresponse Adjustment Household level nonresponse adjustment is more complex than in other surveys given the large number of households which were selected as part of the EMB with unknown eligibility Households who were selected as part of the EMB sample were screened on whether they contain at least one member of a relevant EM group Berthoud et al 2009 Given the low proportion of eligible households in EMB sample it is unrealistic to assume that all nonresponding households would be eligible i e contain at least one EM member To take 29 this into account we modeled eligibility and used this information in household nonresponse adjustments such that households which were more likely to be eligible had higher influence on the nonresponse correction Note that the predicted eligibility multiplied by the design weight is released for all the EMB sample households of unknown eligibility as part of a hhdenus xd This will enable an ad
35. compasses a design weight Wave 1 poststratification and adjustments for non response at each of the Waves 1 to 9 of the BHPS The second component is derived from a model of the propensity to be issued at UKHLS Wave 2 conditional on being enumerated in BHPS Wave 9 and therefore adjusts for all the stages of dropout between BHPS Wave 9 in 1999 and UKHLS Wave 2 in 2010 Model covariates were taken from the Wave 9 household grid and household questionnaire BHPS OSM newborns since Wave 9 England Scotland or Wales or Wave 11 Northern Ireland whose parents are both OSMs were then assigned a base weight equal to the smaller BHPS inclusion weight of their OSM parents in the child s 2010 issued to UKHLS household This reflects the idea that the probability of the child entering the UKHLS sample equals the probability of at least one of his or her parents entering the sample which in turn is equal to or greater than the probability of the parent who has the greatest probability of entering the sample BHPS OSM newborns born to one OSM parent and one TSM parent were assigned a base weight equal to half of the OSM parent s weight in the child s 2010 issued to UKHLS household The division by two reflects the idea that these newborns had double the chance of becoming BHPS OSMs relative to people born to both OSM parents as they would have been included had either their mother s or father s 1991 household been 35 sampled For newborns who were
36. d be included in the questionnaire 4 an incorrect implementation of the check or 5 a problem in exporting and or delivering the data After investigation steps may include correcting the specification data editing reporting the error to NatCen to be fixed in a subsequent delivery and or a quality feedback report suggesting changes to the questionnaire or field practice in subsequent waves Batch specific databases are merged into a single database from which anonymised data is exported for the creation of public use files Data distributions are also checked for theoretical and statistical plausibility This checking is done through direct scrutiny and by analyses which road test the data Documentation of the questionnaires modules and questions The text of the questionnaires in pdf format is part of the documentation provided through the UK Data Archive Questionnaires can also be found http data understandingsociety org uk documentation mainstage questionnaires The documentation is for the mainstage survey household and individual and the adult and youth self completion instruments The instruments are an important source of information about the wording of individual questions who was asked and what questions precede and follow Most of the interview is conducted with a computer assisted personal interview CAPI The CAPI instrument governs the flow of questions and recording of answers but it is not convenient for document
37. d in this model were the same as for the eligibility model and are described in detail below Given that administrative neighbourhood data differs between England Wales Scotland and Northern Ireland a separate model was implemented for each country GPS and EMB response propensity was modeled together which allowed us to model nonresponse within each country separately but the indicator of EMB was retained in the model even if it was not statistically significant Predictors used for eligibility model and household level nonresponse correction come from the following sources e Sampling frame information including such variables as sample month and geographical region e Predicted ethnic density of the postcode sector for five main ethnic groups in England Scotland and Wales as described in Berthoud et al 2009 e Awide range of indicators from Census 2001 and the most updated version of neighbourhood statistics as of summer 2011 linked separately for England Wales Scotland and Northern Ireland see below The household nonresponse correction weight was calculated as the inverse of probability from the above model This weight was multiplied by the household design weight to create the Wave 1 household level weight The design effect was estimated using this weight showing that no truncation was necessary The obtained weight was scaled to a mean of 1 and was named a hhdenus xw Neighbourhood statistics 30 For England and
38. d questionnaire and household grid The base weight for rising 16 year olds correction is continuous enumeration since 1991 b psnen91 lw for the BHPS 1991 main response weight b indin91 lw and is the BHPS 2010 longitudinal enumerated person weight b psnenbh lw see next section below for the BHPS 2001 main response weight b indinO1 lw The main response weight for each rising 16 year old is then scaled by a constant factor so that the ratio of rising 16 year olds to older adults among main questionnaire respondents equals the equivalent proportion among all 34 enumerated respondents The weights x psnen91 lw x psnenO1 Iw x indin91 lw and x indinO1 lw are calculated by multiplying the respective BHPS Wave 18 weight and the adjustment and are scaled to one BHPS cross sectional weights The BHPS cross sectional weights are created as follows we first model the chance of each BHPS OSM being issued into the UKHLS reflected in b psnenbh li then the chance of being in a responding household complete the household grid and the household questionnaire at Wave 2 of the UKHLS conditional on being issued reflected in b psnenbh lw The weight b psnenbh Iw is then extrapolated to TSMs and PSMs through a weight share method to create b psnenbh xw The detailed procedure for creating these weights as well as cross sectional individual response weights is described below The inclusion weight D psnenbh li was calculated separately fo
39. dingsociety org uk design content outlines aspx Changes to the questionnaire Questionnaire changes have been made under certain circumstances At the end of the first six months of data collection in Wave 1 multiple variables were dropped because of the length of the interview e g cutting of the employment history module At the same time other modifications were made e g in question format Notes about these changes have been documented in the variable view of the online documentation system Other fieldwork materials Other fieldwork materials are also on the website http data understandingsociety org uk documentation mainstage fieldwork One example is the Showcards which are used to help respondents with their answers Showcards are referenced in the questionnaire Project Instructions were prepared for interviewer training and to serve as a resource in data collection Documents for communicating with participants are also included on the website In Wave 1 we asked for consent to link to administrative health and education records The information leaflets and consent forms are in this section of the study website The Address Record Form ARF is an important source of information about responding and non responding households It has the call record observations on characteristics of accommodation and households and household outcomes In Wave 1 there are several different versions of the ARF The first distinction is between th
40. don Professor Nick Buck is the principal investigator Fieldwork is conducted by the National Centre for Social Research NatCen with collaboration with the Central Survey Unit of the Northern Ireland Statistics and Research Agency NISRA in Northern Ireland The overall purpose of Understanding Society is to provide high quality longitudinal data about subjects such as health work education income family and social life to help understand the long term effects of social and economic change as well as policy interventions designed to impact upon the general well being of the UK population Route guide for users This release has data for the General Population and the Ethnic Minority boost EMB sample Former participants of the British Household Panel Survey BHPS are part of Understanding Society from Wave 2 http www iser essex ac uk bhps The BHPS is a household panel survey of around 8 000 households in the UK which has completed 18 annual waves of data collection and has been run by ISER since it began in 1991 Data from the BHPS can be obtained from the UK Data Archive SN5151 British Household Panel Study Waves 1 18 1991 2009 http www esds ac uk findingData snDescription asp sn 5151 Data from the Innovation 3 Panel a separate survey intended to support methodological research www understandingsociety org design innovation default aspx Data from the Innovation Panel has been released through the UK Data Archive
41. dp w_hhsize w hsownd w tenure dv w hhtype dv w fihhmngrs dv w emboost w gpcomp w hhresp dv w hhdenus xw Ww psnenus xw psu strata pidp Ww country Description Household identifier Household size House owned or rented Household type Gross household income in past 30 days Ethnic minority boost flag Ppopulation Sample Comparison with EM boost Household response outcome Household cross sectional weight household grid and household interview Primary sampling unit Sampling strata Cross wave person identifier Country or part of the UK 44 w_gor_dv w_ivfio_dv w jbnssec8 dv W Sex w dvage w marstat mpid fpid a nchild dv w jbstat w jbhas w ukborn w fenow a_health w 050 00 Government office region individual response outcome Social class NS SEC Sex Age Legal marital status cross wave identifier of natural mother father number of natural children in household Current economic activity employment status Did paid work last week Born in the UK and UK country of birth Still in further education Highest educational qualification Long standing illness or impairment Current occupation SOC2000 Documentation of derived variables Derived variables are variables that are copied from one file to another for analytic convenience or computed from one or more variables Some are computed by the Blaise CAPI program to control the routing within the questionnaire
42. e General Population Sample GP and the Ethnic Minority Boost Sample EB The versions labeled ARF are longer because they include questions for screening household members for eligibility ARF s labeled 2 or 3 are for addresses with multiple households and or dwelling units Finally there are versions for ARF EB1 Year 1 or Year 2 This change in form was required by the change in selection criteria implemented in Year 2 of Wave 1 see Berthoud et al 2009 for more detail The ARF screening card was a show card used during the screening interviews Additional information about completion of the ARF can be found in the Project Instructions for Interviewers http data understandingsociety org uk documentation mainstage fieldwork docments Sample design The Understanding Society sample consists of a new large General Population Sample plus four other components the Ethnic Minority Boost Sample the General Population Comparison sample the ex BHPS sample and the Innovation Panel sample The design of 18 all five components is described in more detail in an Understanding Society working paper see Lynn 2009 The Innovation Panel is prepared as a separate study which can be accessed via the UK Data Service SN 6849 The General Population Sample is based upon two separate samples of residential addresses one for England Scotland and Wales and one for Northern Ireland The England Scotland and Wales sample is a proportionately strat
43. e detailed variable view of the online documentation Additional codes denote different types of reasons for the lack of a valid response These values have not been specified as missing in Stata or SPSS However these statistical packages have commands to assign values to missing for many variables simultaneously Codes are 9 Missing by error 8 Not applicable to the person or because of routing 7 Proxy respondent The question was not asked of proxy respondents or derived variable cannot be computed for proxy respondents 2 Refused 1 Don t know The meaning of other values is explained with the variable s value labels There may also be notes in the detailed variable view of the online documentation system on the website data understandingsociety org uk documentation mainstage dataset documentation Learning about the study variables There are multiple resources for learning about the study variables in order to plan analyses These include the questionnaires and the module and variable views in the online documentation system Many of the basic non derived variables can be learned about directly from the questionnaires As was shown in Figure 2 the questionnaire has much useful information Please note that in the questionnaire the variable name does not have the wave prefix It also shows the brief variable label text of the question source of the question and value labels Showcards to help the respondent in answerin
44. early months of 2011 The previous interview for most of these households was between September and December 2008 As the gap between the Wave 18 BHPS interview and the Wave 2 Understanding Society interview increased so did the level of untraced movers Refusals as well were generally higher in Great Britain than in Northern Ireland Refusals are expected to be higher at the second wave of a longitudinal study than at subsequent waves The higher than expected refusal rate for the former BHPS sample particularly those in Great Britain may be due to the aforementioned change in the name and logo of the study as well as the change in fieldwork agency and thus for most households a change of interviewer Table 4 below shows the cross sectional response rates for adults in Wave 2 Where a household responded we have an individual level outcome for all adults Where a household did not respond we have assigned the household non response outcome to the adults who were issued to that household From this we can see for example that we were not able to interview 7 229 adults in the UKHLS General Population Sample in Great Britain because they were residing in households who refused to participate at Wave 2 In the Great Britain samples of the former BHPS there is a relatively small group of households who only give telephone interviews On a longitudinal study such as the UKHLS researchers are typically interested in having pairs of observation
45. ed using the above recursive system as starting values in an iterative imputation process In other words the starting values are used to begin a new cycle of imputations where each equation is estimated sequentially but this time using as explanatory variables both X and all the imputed variables Y Yo Y excluding the one used as dependent variable At the end of this new cycle a set of new imputed variables is produced and used to begin a further new cycle of imputations These cycles of imputations are repeated until convergence Notice that in practice some of the variables will be imputed by excluding some of the Xs and Ys variables because it does not always make sense to use all variables as predictors All variables are imputed as reported except for wages and self employment income where we convert amounts reported net to gross where gross is not reported using a deterministic model based on the tax and national insurance system Where net wages and self employment income are not reported we convert from the gross amounts reported or imputed using the same model based on the tax and national insurance system In computing total personal income it is assumed that all other sources are reported gross or are not subject to taxation We will in due course be producing net income estimates Item non response for income variables in the proxy questionnaire The only income variables reported in the proxy questionnaires are the total gross
46. es designed to ensure adequate response and effective data quality ISER has the primary responsibility for design work NatCen manages fieldwork editing and coding and data entry It also advises on the design of all research instruments NISRA collaborates with NatCen and is responsible for fieldwork in Northern Ireland ISER plays a major role in quality control through specification of fieldwork practices survey materials editing and coding requirements and inspecting and analysing weekly fieldwork progress reports This working relationship is reinforced by an agreed set of survey specific procedures to ensure adequate response and effective data quality Full details of these and other technical aspects of the data collection and fieldwork coding and data processing are 5 found in the Technical Reports published each wave on the Understanding Society website see http data understandingsociety org uk Getting ready for fieldwork Prior to the first wave of the main Understanding Society survey there were two small pilot studies and a dress rehearsal A cognitive pilot of 70 individuals was conducted March April 2008 to test screening and other questions relevant to the ethnicity strand A translation pilot was conducted in June 2008 50 interviews were carried out using Bengali and Punjabi translations of the questionnaire to see if there were problems with the operation of the translation program or problems with inter
47. ex marital status ethnicity work e household socio economic variables household size number of children in the household whether there is nobody in the household who speaks English whether the interview had to be translated house type an indicator for whether the person is owner of the house the external condition of the address relative to the others number of bedrooms in the house number of other rooms in the house value of the property for home owners number of cars number of durables log last year s expenditure on domestic fuel e g electricity and gas amount spent on food eaten outside the home in four 40 weeks prior to interview amount spent on food from food shops in four weeks prior to interview weekly rent paid whether the household can keep the accommodation warm enough e government office region indicator for whether the area is a low density area for ethnic minorities Coding Occupational coding for respondent s current or last occupations was carried out by NatCen Coding was carried out on a case by case basis by trained coders with 10 of the coding of SOC and SIC subject to a blind coding check Coding of parental occupations and respondent s first occupation was carried out within ISER using the Computer Assisted Structured Coding Tool CASCOT system developed by Peter Elias As a result of the six figure codes attached via CASCOT matching of the 1990 SOC coding with previous occupational class
48. g are also marked as part of the questionnaire You can go back and forth from the question view to the variable view 43 Identifiers and useful variables Households are identified by w_hidp a wave specific variable with a different prefix for each wave It can be used to link information about a household from different records within a wave but cannot be used to link information across waves Since the composition of households change between waves the data do not include a longitudinal household identifier Individuals are identified by the personal identifier pidp which is consistent in all waves and can be used to link information about a person from different records belonging to one wave or to link information from different waves Individuals are also identified by w_pno the person number within the household The combination of w_hidp and w_pno is unique for each individual Table 9 lists some variables commonly used in analysis and may help the analyst to begin planning Recall that the variables with the prefix w_ have the values for that wave There is also the file XWAVEDAT which has variables with stable values Variables in that file do not have the wave prefix Analysts should also remember to consult the section on specifying the complex sampling variables from Section 2 Sample design variables and analysis and on weighting from Section 2 Weighting adjustments Table 9 Some useful variables Variable w_hi
49. g confidentiality In preparing the data for the release we have taken steps to maintain the confidentiality of responses These include not releasing the full date of birth and not releasing detailed geographic identifiers Houshold income has been top coded Open or narrative text e g names of schools or employers has not been released since it may indirectly identify individuals A Special License version of the data will be released through the UK Data Archive The study has a Data Access Committee to take decisions on applications requesting access to electronic data and biological samples from Understanding Society Its aim is to allow important research to proceed while minimising risks particularly to Study participants Paradata Some paradata additional data collected about the interview process is available These consist of call records timings data and other information collected by the interviewers during the interview The W CALLREC data file has information the number of calls made as well as the issue number time and date and the outcome of each call Information on the date of receipt of the case and the interviewer associated with each issue as well as the outcome at the end of each issue period is available in the file W ISSUE In addition to this information collected in the address response form ARF by interviewers while contacting each household and asking household members to participate in the survey is available
50. g that would indicate that this information relates to the spouse or partner renpfix a sp rename the spouse partner pno variable to the respondent pno variable as this will be used to match on to the respondent information Then sort and save the data rename sp hgpart a pno rename sp hidp a hidp drop sp pno sorta hidp a pno save spousepartner replace Again open the data with information on all persons in responding households use a hidp a pno a hgparta sexa dvage using a indall ip if a_hgpart gt 0 clear rename the prefix a to something that would indicate that this information relates to the respondent as we want to match hidp and a_pno rename hidp and r pno back to these rename r hidp a hidp rename r pno a pno 48 Now sort and merge with the spouse partner file sort a_hidp a_pno merge 1 1 a_hidp a_pno using spousepartner drop merge save final3 replace erase spousepartner dta Example 4 Using the egoalt file to create household composition variables In this example we will create a variable that measures the number of siblings in the household using the egoalt file use b hidp b epno b relationship using b ip clear create a variable that counts the number of siblings in the household bysort b hidp b epno egen nsiblings sum b_relationship gt 14 amp b relationship 17 lab var nsiblings number of siblings in household keep o
51. ght for BHPS sample They represent the probability of a BHPS sample member being included in Understanding Society i e equivalent to a design weight but including non response over 18 years or so It is not an analysis weight Examples a indinus xw is the cross sectional analysis weight for individual interview data from Wave 1 representing the population of persons aged 16 or older b indscus lw is the longitudinal analysis weight for individual self completion interviews from Wave 1 and Wave 2 representing the adult population who continuously lived in UK at the times of Wave 1 and 2 Technical details In this section we describe in turn how the weights were derived for UKHLS GPS and EMB wave 1 weight e UKHLS GPS and EMB longitudinal weights e UKHLS GPS EMB cross sectional weights after Wave 1 28 e longitudinal weights e cross sectional weights UKHLS Wave 1 weights The Wave 1 household level weights consist of two components a design weight and nonresponse adjustment for household level nonresponse Wave 1 individual level weights consist of four components the design weight nonresponse adjustment for household level nonresponse individual level within household nonresponse and post stratification to population characteristics Each of the components is explained below Design weight The design weight corrects for unequal probability of selection at a number of levels The household le
52. hold level covariates taken from responses to the UKHLS Wave 2 household grid and household questionnaire The BHPS cross sectional household weight b hhdenbh xw is set equal to the minimum cross sectional person enumerated weight psnenbh xw amongst adults in the household Each weight has been scaled to have a mean of one amongst cases eligible to receive the weight Imputation of income variables Understanding Society collects detailed information each wave on personal income All individuals aged 16 or more are asked to report wages self employment earnings second job earnings interest and dividends pensions National Insurance state retirement pension pension from a previous employer pension from a spouse s previous employer private pension annuity widow s or war widow s pension widowed mother s allowance or widowed pension e benefits severe disablement allowance disability living allowance war disablement pension attendance allowance carer s allowance incapacity benefit income support job seeker s allowance national insurance credits child benefit child tax credit working tax credit maternity allowance housing benefit council tax benefit foster allowance guardian allowance rent rebate rate rebate employment and support allowance respond to to work credit sickness and accident insurance in work credit for lone parents and pension credit and 36 e other income sources educational gran
53. hs of first year Discrimination X EMB GPC LDA and job variables Parents and Children Family Networks X responsible mother and responsible father of children X Remittances X EMB GPC LDA Harassment X EMB GPC LDA Environmental X Behaviour Consents for linkage to health and education administrative records X EMB ethnic minority boost GPC General Population Comparison LDA Low density area Own first job will be asked of OSMs and new entrants in Wave 4 The self completion questionnaires are not divided into modules but Table 7 summarizes the content in waves 2 and 1 Table 7 Summary of Adult Self Completion Questionnaires in Waves 2 and 1 Wave 2 Wave 1 GHQ 12 X X Satisfaction life and other X X Alcohol consumption X Control X Positive and negative social X support SF 12 X in adult interview Health and Disability module Gender role opinions X Identity X Sleep X Environmental attitudes and X 17 beliefs Neighbourhood belonging and participation Trust Short Warwick Edinburgh Mental Well Being Scale Attitudes to risk Partnership relationship quality activities happiness The content of the Youth self completion instruments is summarized in pages 6 7 of the long term content plan http www understan
54. ifications is now possible in addition special algorithms within CASCOT allow the re coding of SOC codes into SEG RGSC Goldthorpe Hope Goldthorpe Cambridge Scale and ILOISCO 88 Several questions e g country of birth religion political party national identity and citizenship had an other please specify option These responses were coded using an automated process Coding was also done for an open ended question We ve asked you a lot of questions but we also want to know what has happened in your own life that has been especially important to you Can you please tell me anything that has happened to you or your family over the past year that has stood out as important The respondent could give up to four answers The answers were recorded verbatim and manually coded for type of event and its subject 3 File and variable information The data release consists of multiple files in SPSS or Stata formats distributed by the UK Data Service The list of files and their descriptors can be seen in the online documentation system http data understandingsociety org uk documentation mainstage dataset documentation Information about the BHPS Sample Component This release of Wave 2 data contains two Understanding Society samples 1 The General Population and Ethnic Minority Boost sample for Waves 1 and 2 2 The sample from the former British Household Panel Study BHPS Both samples can be used for cross sectional and
55. ified equal probability clustered sample of addresses selected from the Postcode Address File The Northern Ireland sample is an unclustered systematic random sample of addresses selected from the Land and Property Services Agency list of domestic addresses General Population Sample component The sample for England Scotland and Wales was selected in two stages The first stage was to select a sample of postcode sectors to serve as primary sampling units The second stage was to select addresses within each sampled sector Prior to selection any postcode sector with fewer than 500 residential addresses was first grouped with an adjacent sector and thereafter treated as a single sector The list of all sectors was then sorted into twelve geographical strata consisting of ten regions in England plus Scotland and Wales as separate strata Within each of the twelve strata sectors were sorted into three sub strata based upon the proportion of household reference persons classified as non manual workers based on 2001 Census data Within each of the 36 sub strata sectors were then sorted into three further sub divisions based on population density households per hectare and within each of the 108 resultant sub divisions sectors were listed in order of ethnic minority density From the sorted list a systematic random sample of 2 640 sectors was selected with probability proportional to the number of residential addresses in the sector These sectors
56. in 2001 For further details of the BHPS sample see section IV of the BHPS User Guide http www iser essex ac uk bhps documentation vola vola html Sample status and following rules There are three possible sample statuses Original Sample Members OSMs Temporary Sample Members TSMs and Permanent Sample members PSMs The definitions are as follows Original Sample Members OSMs All members of Understanding Society General Population Sample households enumerated at Wave 1 including absent household members and those living in institutions who would otherwise be resident are Original Sample Members OSMs All ethnic minority members of an enumerated household eligible for inclusion in the Ethnic Minority Boost sample are OSMs In the Innovation Panel all members of households enumerated at Wave 1 and refreshment sample households enumerated at Wave 4 are OSMs In all of these samples any child born to an OSM mother after Wave 1 and observed to be co resident with the mother at the survey wave following the child s birth is an OSM In the formerBHPS sample OSMs are those who were enumerated at the first wave of the sample from which they come Wave 1 for the original sample Wave 9 for the Scotland and Wales boost samples Wave 11 for Northern Ireland or who were subsequently born to an OSM mother or father or both From Wave 2 onwards of Understanding Society in the former BHPS sample as for the rest of the Understanding Societ
57. ing the advice provided below The weighting strategy is described in Lynn and Kaminska 2010 The first part of this section covers the purpose of the weights and how to use the naming conventions for the weight variables to interpret and select the different weight variables from among a complex assortment This is followed by the technical details of how weights were calculated If your aim is to generalise to the UK population do not conduct unweighted analyses For advanced users who want to model nonresponse in their own way we provide design 23 weights see below which adjust the sample for unequal selection probabilities Note that adjusting for the first wave nonresponse is different from adjusting for attrition and requires variables which have values for both responding households and never responding households In this release we do not provide a weight for combining the former BHPS sample component with the General Population and Ethnic Minority Boost sample components At this point we recommend that analyses be carried out separately for the two samples Note that the two samples represent slightly different populations as the BHPS does not represent people who were not resident in the UK in 1991 or their descendants Selecting the correct weight for your analysis Given the complexity and multi purpose nature of the UKHLS design we provide multiple weights to meet different needs of users The weight for your analysis
58. ked at the Super Output Area SOA level and was obtained from http www ninis nisra gov uk Examples of predictors obtained from Census 2001 at the SOA level include the average hours worked by residents the average age of residents percentages of residents with different level of qualifications with different employment statuses and with different types of marital status among others The predictors also include 2007 2009 information on multiple deprivation indexes Note that using Understanding Society analysis weights all but design weights adjusts for household nonresponse bias in any estimate to the extent it is related to the above mentioned variables Enumerated Individual Weight The weight for analysis of enumerated individuals a_psnenus xw is not equivalent to the household weight for all household members as often happens in other household studies This is because we have TSMs in Wave 1 who are not ethnic members selected into EMB part of the sample Thus the individual level design weight is not equal to the household level design weight for individuals in households containing a mix of EM and non EM persons The weight for the analysis of enumerated individuals is calculated as the product of individual level design weight a_psnenus_xd and the household level nonresponse correction described above The design effect was tested showing that no truncation was necessary Weighted sample distributions were then compared to O
59. lable under Special License conditions http www esds ac uk findingData ukhIsSL asp Notifications to ISER can be sent to info understandingsociety org uk 52 5 Citations and acknowledgements Users should acknowledge both the UKDA and the Institute for Social and Economic Research in any publications arising from analysis of the data Citation of the data University of Essex Institute for Social and Economic Research and National Centre for Social Research Understanding Society Wave 1 2 2009 201 1 computer file 4 Edition Colchester Essex UK Data Archive distributor December 2012 SN 6614 http dx doi org 10 5255 UKDA SN 661 4 4 Citation of the User Manual McFall Stephanie ed 2012 Understanding Society UK Household Longitudinal Study Wave 1 2 2009 2011 User Manual Colchester University of Essex People who participated in writing sections of the documentation included Jon Burton Peter Lynn Olena Kaminska Gundi Knies Randy Banks Cheti Nicolletti Laura Fumagalli Jakob Petersen and Nick Buck Many people participated in preparing and processing the questionnaires and data From the information technology side we recognize the contributions of Paul Groves Paul Siddall Geoffrey Angel Tom Butler Jeannette Chin Elaine Prentice Lane and Catherine Yuen From the survey research team we recognize Noah Uhrig Sarah Budd and Emily Dix A small group was active in contributing code for derived v
60. longitudinal analyses For both these purposes they will need to be analysed separately because of their different sampling histories Separate weights are provided for the two samples as described in Section 2 Weighting adjustments 41 The cases in the two samples can be distinguished using the variable b_memorig for person level files and b_hhorig for household level files These variables also allow the identification of different components of the BHPS sample see below The questionnaires used for the two samples are the same There are however a few differences in the data collected One important issue is that the date of previous interview for GPS sample members who were interviewed at the previous was approximately 12 months earlier while for the former BHPS sample the gap was between 13 and 27 months for sample members interviewed at Wave 18 of BHPS This means that the reference period for history of events since the last interview will be longer for the BHPS sample For longitudinal analysis of the GPS sample cases may be matched to Wave 1 data available as part of this release from the UK Data Service using the variable pidp the Understanding Society cross wave person identifier However for the BHPS sample a different identifier will need to be used the variable pid which is the BHPS cross wave person identifier The pid identifier is available in all person level files in the Understanding Society Wave 2 release a
61. mother and observed to be co resident with the father or any other OSM at the survey wave following the child s birth TSMs remain eligible for interview as long as co resident in an OSM PSM household TSMs who are not co resident in an OSM PSM household are not followed and become ineligible for interview TSMs are identified as re joiners if they are subsequently found in an OSM PSM household and then become eligible for interview Permanent Sample Members PSMs PSMs are TSMs who are followed for interview after they no longer live with an OSM This is done for substantive research reasons because of the additional contextual information they may provide for the analysis of OSMs At present there is only one category of PSM but others may be defined in the future Any TSM father of an OSM child born after Wave 1 and observed to be co resident with the child at the survey wave following the child s birth is a PSM PSMs remain potentially eligible for interview for the life of survey Sample design variables and analysis As the sample design involves stratification clustering and weighting these design features affect standard errors and should therefore be taken into account in analysis Appropriate variables are provided to allow the analyst to do this The weighting variables are described in a separate section Here we describe the stratification and clustering variables psu This is an indicator of the primary sampling unit PSU
62. mpletion interview For example if in one model from Wave 1 you use questions from the proxy and full interview as well as from the self completion then the correct weight will be a indscus xw 24 the weight for the self completion questionnaire as its level 1 is lower than the level for proxy and full interview 3 Table 8 List of weight variables by analysis level wave and data source Analysis level eee source Analysis Weight Household lans NE and or household interview a hhdenus xw Household grid and or household household interview b hhdenus xw Household grid and or household interview BHPS b hhdenbh xw Household grid and or household individual interview a psnenus xw Household grid and or household individual interview b psnenus xw Household grid and or household individual interview BHPS b psnenbh xw Household grid and or household individual interview b psnenus lw 2008 Household grid and or household individual interview BHPS GB 1991 b psnen91 lw 2001 2008 Household grid and or household individual amp 2 interview BHPS UK 2001 b psnenO1 lw uu pcr ee main and proxy interview a 22888 Adult main and proxy interview individual BHPS b_indpxbh_xw b qure MN NM D oe household household Britain amp 2 2001 2008 individual Adult self completion b indscus Iw individual Youth self completion individual Youth self completion b y
63. nd in the 18 wave BHPS longitudinal data set available separately from UK Data Service SN5151 British Household Panel Study Waves 1 18 1991 2009 http www esds ac uk findingData snDescription asp sn 5151 While the great majority of BHPS sample cases who were interviewed in Understanding Society Wave 2 were previously interviewed at Wave 18 in 2008 9 there are a number who were last interviewed at an earlier wave Information about the response status of BHPS sample members at each of the 18 waves is contained in the BHPS file XWAVEID The BHPS data set also contains a file called XWAVEDAT which contains the values for stable variables e g ethnic group parent social class etc Because of some differences in variable definition this information has not been copied across to the new Understanding Society file also called XWAVEDAT However in most cases values of these variables can be obtained by matching to the BHPS file We hope to produce a harmonized version at subsequent release In matching to earlier waves of BHPS data it is important to be aware that variable names in the BHPS data set have slightly different formats e they are limited to eight characters e there is no underscore separating the wave prefix from the main part of the name e derived variables imputation flags weights and other special variables are not distinguished by _ dv if suffixes However most questionnaire variables which are carried in both sur
64. ne observation per person bysort b hidp b epno keep if _n sortb hidp b epno save final4 replace Now this information can be merged with any individual level file Example 5 Merging individual files across waves into long format To match individual level files across two waves into a long format do the following for more waves add wave specific prefix in the foreach statement foreach w ina b open the individual level file 49 use pidp w _jbhas using w_indresp_ip clear drop the wave prefix from all variables renpfix w _ create a wave variable gen wave sstrpos ab w save one file for each wave save temp w replace open the file for the first wave wave a use tempa clear foreach w in b append the files for second wave onwards append using temp w save the long file save final5 replace erase temporary files foreach w erase temp w dta Example 6 Merging individual files across waves into wide format To match individual level files across two waves into a wide format do the following for more waves add wave specific prefix in the foreach statement 50 use pidp a jbhas using a indresp ip clear sort pidp save temp replace foreach w in b use pidp w jbhas using w indresp ip clear sort pidp merge 1 1 pidp using temp drop merge sort pidp save temp replace save final6 replace erase temp dta Preservin
65. nic minority members living with TSM ethnic minority members Giving a cross sectional weight of 0 to Wave 1 TSMs maintains the balance of the whole sample These cross sectional enumerated individual weights then serve as the base for the other cross sectional individual level weights each of which main main or proxy self completion youth involves an additional adjustment for non response to the relevant instrument conditional on enumeration The non response models are therefore based on all eligible persons enumerated at Wave 2 including TSMs and those OSMs who did not respond to the respective instrument at Wave 1 with covariates taken from responses to the UKHLS Wave 2 household grid and household questionnaire The cross sectional weights for households b_hhdenus_xw are set equal to the minimum nonzero longitudinal enumerated person weight b psnenus amongst adults in the household reflecting the idea that the probability of observing the household is equal to or greater than the probability of observing the person in the household who has the greatest probability of being observed 1 Note that there is no cross sectional weight for the Extra 5 minutes questions as at Wave 2 these were only asked of sample members who had completed the main interview at Wave 1 Thus the wave 2 longitudinal weight should be used for Wave 2 cross sectional analysis 33 BHPS longitudinal weights Four weights will be continued f
66. nto GPS or EMB that the response propensity is assumed to not depend whether respondents received the Extra five minutes or not and that conditional on age present in the model the response to self completion is assumed to have the same predictors for adults and youth this assumption allowed modelling the response in each country separately which wouldn t otherwise be possible for youth sample The individual level response conditional on household response was modeled using backward stepwise logistic regression separately for England Wales Scotland and Northern Ireland The four models were implemented for each of the three levels described above The predictors used in the models include all the predictors used for the household level nonresponse models and individual and household level variables obtained from the household questionnaire such as age and gender marital and employment status household size and presence of children in the household as well as household expenditure on food and food outside consideration of use of environmental energy among others The individual level non response adjustment was obtained as the inverse of the predicted probability and was then multiplied by the relevant either individual or Extra five minutes design weight and by the household nonresponse correction No truncation was deemed necessary as there were no extreme values substantially impacting design effects The poststratification
67. odation outright is it being bought with a mortgage is it rented or does it come rent free Interviewer Instruction The text is what the interviewer reads F9 FOR HELP Options 1 Owned outright 2 Owned being bought on mortgage 3 Shared ownership part owned part rented Value labels 4 Rented 5 Rent free 97 Other Use Ask Hsownd Modules This question comes from Wave 1 ModuleHousehold w1 Household Questionnaire Household Questionnaire module Figure 2 shows a marked up sample page from the individual interview The question is more complex The question is asked about each natural or biological child so multiple variables are associated with the question for each natural child The variables are located in the data file A NATCHILD which has one record for each natural child 14 Figure 2 Mark up of question with looping from individual questionnaire Brfed Breastfeed Variable name amp Variable label Source UKHLS es Question may be asked multiple t Did you breastfeed even if only for a short time may ed 4 About each resident child Options 1 Yes 2 No Values labels 3 Currently breastfeeding applies for children 5 in household only Use Ask BrFed Modules 2 ModuleFertilityhistory_w1 Fertility history module Question is from Wave 1 Fertility history module Sections Section individual interview Universe It LNPrnt gt 1 Lert 7 Parent of biological child Who is eligible to be asked
68. om the previous wave Example code for matching files We are including six examples of common data management tasks useful in analysing the data Each task is illustrated with code for Stata Because Stata is case sensitive we have not displayed file and variable names in upper case but in lower case Statements beginning with are comments The six tasks include e Distributing household level information to individual level Summarising individual level information at the household level e Matching individuals within a household e Using the egoalt file to create household composition variables e Merging individual files across waves into long format e Merging individual files across waves into wide format Example 1 Distributing household level information to individual level In this example we will distribute household level information to individuals in those households We can do this by merging household level file such as w household with an individual level file such as w indresp within the same wave open the household level file use a hidp a hhsize using a hhresp ip clear sort it on the household identifier w hidp sorta hidp save this temporary file save hhinfo replace open the individual level file use pidp a a marstat using a indresp ip clear sort it on the household identifier w hidp sorta hidp merge it with the earlier saved file on w hidp The output shows how man
69. on is released through the UKDA we encourage users to consult the Understanding Society webpage The documentation will develop over time We plan to be developing specific guides about major content areas such as the biomeasures or cognitive measures and guides for issues that are frequently problematic for users such as selection of appropriate weights Most of the Wave 1 has been released according to the conditions of the regular UKDA End User License https www esds ac uk aandp access licence asp A version of the Wave 1 to Wave 2 data has been released under conditions of the Special Licence SL SN 6931 Special License datasets are anonymised but contain more detailed information than End User Licence EUL data The UKDA requires users to complete a set of forms with such detail as the intended use of the data Researchers are asked to report publications resulting from the data Related Understanding Society releases are being prepared One is a set of data products with information to link Understanding Society survey data with geographic units including Local Authority Districts Area Classification for Output Areas Travel to Work Areas Westminister Parliamentary Constituencies Rural urban Indicators Local Education Authorities and Primary Care Trusts For further information about these geographic units see Office for National Statistics 2010 or the working paper on this topic Rabe 2011 The geographical look up tables are avai
70. ormation about the survey and thanking respondents for participating A minimum of six calls is made at each sampled address before it is considered a non contact Interviewers are encouraged to make further calls if possible If there was a potential for success a special conversion letter is sent to households which had refused to participate or had not been contacted Post interview quality control is carried out with a telephone recall on 10 of all completed interviews Interviewers upload their work daily including information about all the calls they have made whether or not there was any response This information is collated by NatCen to construct a weekly field progress monitor report for ISER Panel membership and panel maintenance The rules for following individual respondents over time stem from the composition of the household Individuals found at selected households in the first wave were designated as Original Sample Members OSM We attempt to maintain OSM respondents as part of the sample as long as they live in the UK In addition births to an OSM mother are also classified as OSM Individuals joining the household of an OSM after enumeration of the household at Wave 1 are Temporary Sample Members TSM One deviation from this is for individuals who were not an ethnic minority within the households selected as the ethnic minority boost sample At Wave 1 these individuals were classified as TSMs We attempt to interview TSM
71. r a Northern Ireland and b England Scotland and Wales For each it has two components For Northern Ireland the first component consists of the BHPS Wave 11 cross sectional weight as this is the wave at which Northern Ireland first entered the BHPS This component encompasses a design weight poststratification and an adjustment for Wave 11 nonresponse The second component is derived from a model of the propensity to be issued at UKHLS Wave 2 conditional on being enumerated in BHPS Wave 11 This therefore adjusts for all the stages of dropout between BHPS Wave 11 in 2001 and UKHLS Wave 2 in 2010 Model covariates were taken from the Wave 11 household grid and household questionnaire This propensity was modelled as a single step from 2001 to 2010 because across wave response patterns varied greatly between the sample members There is no single BHPS wave since Wave 11 at which all the Northern Ireland sample members of those issued to UKHLS responded and therefore no other survey instrument that can provide model covariates for all relevant sample members Similarly for England Scotland and Wales the first component consists of the BHPS Wave 9 longitudinal weight as this is the wave at which the Scotland and Wales boost samples were added so all of the members of those samples who entered UKHLS were enumerated at that wave as were the vast majority of members of the original BHPS Wave 1 sample who entered the UKHLS This component therefore en
72. reflects the survey instrument which is the source of the data being used in the analysis and the analysis level household or individual Each weight has been scaled to have a mean of one amongst cases eligible to receive the weight All weights follow a naming convention designed to help users to pick the correct weight The name of each weight reflects the wave for which the weight is calculated level of analysis data source and its nature design weight cross sectional analysis weight or longitudinal analysis weight The rules are described in the Naming Conventions for Weighting Variables section below If your analysis uses only data from Wave 2 select the xw cross sectional version of the weight This weight is defined for all sample members who responded to the relevant survey instrument at Wave 2 If your analysis uses data from both Wave 1 and Wave 2 select the lw longitudinal version of the weight This weight is defined for sample members who responded to the relevant survey instrument at both waves For individual level analysis you may want to combine information from different questionnaire sources In this situation please select the weight suitable for the lowest level according to the hierarchy below Level of Analysis Questions available for household level all enumerated 4 individuals 3 Adult proxy and main interview 2 Adult main interview only no proxy 1 Adult or youth self co
73. rom BHPS Their variable names are changed The corresponding weight variables are e Xlewght now called x psnen91 e Xlewtuk1 now called x psnenO1 e Xlrght now called x indin91 lw and e Xlrwtuk1 now called x indin01 lw where x represents the most recent UKHLS wave These weights are based on Wave 18 BHPS longitudinal weights which account for the first wave household nonresponse the first wave within household individual nonresponse to enumeration or to an individual main questionnaire respectively and for individual nonresponse between the first wave and Wave 18 of BHPS The base weights which reflect continuous enumeration rlewght a BHPS variable name and continuous response to the main questionnaire rlrght a BHPS variable name since 1991 are used for creating weights for longitudinal analysis starting 1991 Note such an analysis excludes Northern Ireland as it was added to BHPS in 2001 and will also exclude the Scotland and Wales boost samples that were added in 1999 Similarly the base weights which reflect continuous enumeration rlewtuk1 a BHPS variable name and continuous response to main questionnaire rlrwtuk1 a BHPS variable name since 2001 are used for creating weights for longitudinal analysis starting in 2001 Analysis using these weights will include all the BHPS samples For more information on the BHPS weight calculation please refer to BHPS documentation Taylor 2010 For each of the Wave 18 weight
74. s an additional adjustment is applied to correct for attrition between Wave 18 of the BHPS and Wave 2 of Understanding Society when the BHPS joined Understanding Society The adjustment is the reverse of the estimated probabilities of participation enumeration or response to main questionnaire based on logistic regressions predicting participation at Wave 2 of UKHLS conditional on participation at Wave 18 of BHPS The covariates used in the model predicting enumeration are from the BHPS Wave 18 household grid and household questionnaire The same covariates plus covariates from the Wave 18 main questionnaire are used for predicting response to the UKHLS Wave 2 main questionnaire Enumeration weights for newborn babies biological step or natural born to an OSM mother since the time of the BHPS wave 18 interview are equal to their mother s enumeration weight For rising 16 year olds OSMs who turned 16 between the time of the BHPS Wave 18 interview and the UKHLS Wave 2 interview and who could therefore be aged 16 17 or even 18 at the time of UKHLS Wave 2 main response weights consist of the relevant longitudinal enumerated person weight with an adjustment for the probability of main response at Wave 2 conditional on enumeration at Wave 2 The adjustment is the inverse of the response propensity predicted by a separate logistic regression model based just upon all adults and inferred to rising 16 year olds using covariates from the Wave 2 househol
75. s of ethnic group membership Members of the General Population Comparison sample are a random subsample of the General Population Sample component and they should be included in analyses of the General Population Sample component 19 Ethnic Minority Boost Sample The Ethnic Minority Boost Sample was designed to provide at least 1 000 adults from each of five groups Indian Pakistan Bangladesh Caribbeans and Africans The initial step was identifying postal sectors with relatively high proportions of relevant ethnic minority groups based upon 2001 Census data and more recent Annual Population Survey data The set of 3 145 sectors constituted approximately 35 of the sectors in Great Britain and covered between 82 and 93 of the population of the five ethnic minority groups The 3 145 sectors were sorted into four strata based on the expected number of ethnic minority households that would be identified by the sampling and screening procedures see Berthoud et al 2009 for details All sectors were included for the stratum where a yield of three or more households was expected In the other three strata sectors were sub sampled at rates of 1 in 4 1 in 8 or 1 in 16 respectively This was done to constrain the number of sectors that might have just one or two eligible sample households or even none The total number of postal sectors selected for inclusion in the ethnic minority boost sample was 771 Of these 6 were in Scotland 7 were
76. s on the same individual to investigate individual level change over time Table 5 below takes as the baseline all those who gave a full interview at the previous wave and shows their outcome at Wave 2 For the former BHPS sample the previous wave was Wave 18 LIB Wave 10 LIS and LIW or Wave 8 NIHPS all collected in 2008 Once more we see that there is a higher re interview rate in the Northern Ireland samples than in Great Britain The lowest re interview rate is in the ethnic minority boost sample largely due to a higher level of non contacted households or households who moved but could not be traced Interestingly the re interview rate was higher in the General Population GB sample than in the three samples that made up the former BHPS GB samples Overall in the Waves 1 and 2 data pairs of observations are available for 45 836 adults If proxy and telephone interviews are included this increases to 47 282 adults For more detail please see the working paper on non response and attrition Lynn et al 2012 10 Table 4 Cross sectional individual adult response rates by sample origin UKHLS GP sample EMB Former British Household Panel Survey UKHLS UKHLS Living in Living in Living in GB NI Britain Beotlarid Walos WIES Total Full interview 32 381 1 770 4 978 6 140 1 461 1 651 2 008 50 389 60 8 62 3 46 3 61 6 58 1 59 6 71 7 59 4 Proxy interview 2 722 87 615 253 49 86
77. sponding households Responding households are households for which the household questionnaire and information on the household composition structure household grid module are available We suggest that the user take account of household non response via weighted estimates described in Section 2 Weighting adjustments For individuals who respond to the individual questionnaire but do not provide answers to all income questions item non response we impute the following personal income variables wages self employment earnings second job earnings interests and dividends pensions benefits and other income sources For individuals for whom a proxy questionnaire is available we impute total earnings and total income whenever missing The proxy questionnaire is a short version of the individual questionnaire with questions on total earnings and total income as well as other variables Finally for individuals in responding households for whom neither the personal nor the proxy questionnaire is available we impute only the total personal income This is not directly included in the data set but is used in the imputation of total household income Based on these imputations we can compute total personal and household income for all individuals belonging to responding households 37 For each income variable for which amounts are imputed there is a separate imputation flag variable with a suffix if instead of dv indicating whether the
78. statbase product asp vlnk 15106 Rabe B 2011 Geographic identifiers in Understanding Society Understanding Society Working Paper 2011 01 Colcheser University of Essex http research understandingsociety org uk publications working paper 201 1 01 Ragunathan E T Lepkowski J M van Hoewyk J and Solemberger P 2001 A Multivariate technique for multiply imputing missing values using a sequence of regression models Survey Methodology 27 1 pp 85 95 Rubin D B 1987 Multiple imputation for nonresponse in surveys New York Wiley Schafer J 1997 Analysis of Incomplete Multivariate Data Chapman amp Hall London Taylor M F ed 2010 British Household Panel Survey User Manual Volume A Introduction Technical Report and Appendices Colchester Universtiy of Essex van Buuren S H C Boshuizen and D L Knook 1999 Multiple imputation of missing blood pressure covariates in survival analysis Statistics in Medicine 18 681 694 54
79. survey It tests varying measurement issues and its instruments are somewhat different from the mainstage survey The IP can be accessed through the UK Data Service SN 6849 Data collection and response outcomes Overview Figure 1 shows the timing of data collection for the data included in this release and for the previous wave of data collection for each of the two samples in this release namely wave 1 of Understanding Society and wave 18 of the BHPS The BHPS may be accessed via SN 5151 from the UK Data Service Data collection for a single wave is scheduled across 24 months There is some variation in that pattern The data collection for Northern Ireland and the former BHPS sample component takes place in the first 12 months of the wave Most of the data collection is conducted face to face via computer aided personal interview CAPI There are also self completion instruments for youth and adults Figure 1 Timing of Data Collection 2008 2009 2010 2011 Q1 Q2 Q3 04 Q1 Q2 03 Q4 Q1 02 03 Q4 Q1 Q2 Q3 Q4 BHPS Wave 18 Wave 1 year 1 Wave 1 year 2 Wave 2 year 1 Wave 2 year 2 Q Quarter Data Collection The players who does what ISER together with NatCen and the Central Survey Unit of Northern Ireland Statistics and Research Agency NISRA work closely together on all aspects of data collection implementing an agreed set of survey procedur
80. t trade union and friendly society payment maintenance or alimony payments from a family member not living together amount for rent from boarders or lodgers rent from any other property These personal income variables can be summed to obtain the total personal income Total household income can be computed from the personal total incomes of all household members Some of the income components can be missing More precisely there can be three types of missing cases 1 item non response when individuals respond to the individual questionnaire but do not answer to some or all the questions on income components 2 individual non response when individuals fail to respond to the individual questionnaire 3 household non response when there is neither a household nor the individual questionnaire response For example at Wave 1 we have 59 466 individuals for whom at least the household questionnaire is available and among these individuals 80 3 provided a personal interview 5 5 have a proxy interview whereas 14 2 had neither a proxy nor a personal interview The item non response rate for individuals who provided an individual questionnaire varies across income variables It goes from a maximum of about 50 for self employment earnings to zero for some of the benefit variables and it is generally below 20 for the remaining income variables What do we impute In Understanding Society we do not impute income variables for non re
81. t aic Dd 53 Understanding Society UK Household Longitudinal Study Wave 1 2 2009 2011 User Manual 1 Introduction Overview of study Understanding Society the UK Household Longitudinal Study UKHLS is a longitudinal survey of the members of approximately 40 000 households in the United Kingdom England Scotland Wales and Northern Ireland Households recruited at the first round of data collection are visited one year later to collect information on changes to their household and individual circumstances Interviews are carried out face to face in respondents homes by trained interviewers Wave 1 data collection took place between January 2009 and January 2011 Wave 2 took place between January 2010 and April 2012 Understanding Society is funded by the Economic and Social Research Council and with funding from multiple government departments the Department for Work and Pensions the Department for Education the Department for Transport the Department for Culture Media and Sport the Department for Communities and Local Government the Department of Health the Scottish Government the Welsh Assembly Government the Northern Ireland Executive the Department for Environment Food and Rural Affairs and the Food Standards Agency The scientific leadership team is from the Institute for Social and Economic Research ISER of the University of Essex the University of Warwick and the Institute of Education University of Lon
82. thscus xw individual Youth self completion BHPS b ythscbh xw For advanced users only Wave 1 household design weight Design weight Extra 5 minutes design weight BHPS inclusion weight for OSMs b psnenbh li issued into UKHLS BHPS 2010 longitudinal b psnenbh lw enumerated person weight Not using weights Note that an unweighted analysis does not reflect the population structure correctly unless the assumptions below are true It is suggested that researchers publishing or presenting unweighted estimates make these assumptions explicit a ythscus xw household individual individual individual individual If no weighting is used an analysis of the UKHLS assumes that all estimates of interest are the same in Northern Ireland as in the rest of the UK that people who live at an address with more than three dwellings or more than three households are the same as those who don t e that people who responded at Wave 1 are the same with respect to your estimates as those who did not that people who continued to respond at Wave 2 are the same 26 as those who did not and that people who responded to each particular instrument used in the analysis individual interview self completion questionnaire etc are the same as those who did not see Lynn et al 2012 An unweighted analysis of the former BHPS sample assumes that all estimates of interest are the same in each of England Scotland Wales and Northern
83. ties and differential Wave 1 response at both household and individual levels and at the individual level included a poststratification adjustment to mid year population estimates by age sex and region The Wave 1 weight was then multiplied by the Wave 2 adjustment to create the Wave 2 longitudinal weight Newborns born to an OSM mother since the Wave 1 interview received the longitudinal enumerated person weight of their mother reflecting the idea that the probability of observing the newborn is equal to the probability of observing the mother The principle behind the longitudinal weights is that they are defined for each person who is observed at all of the relevant waves for which they were eligible For this reason newborns observed at Wave 2 receive a Wave 1 Wave 2 longitudinal weight as they were enumerated at Wave 2 the only wave for which they were eligible UKHLS cross sectional weights after Wave 1 The cross sectional enumerated individual weights are based on the longitudinal enumerated individual weights which are shared to temporary sample members TSMs and permanent sample members PSMs who entered the sample at Wave 2 through a weight share method Note that only new TSMs and PSMs entering the study after Wave 1 receive a shared weight TSMs who were present in Wave 1 in the EMB sample are given a cross sectional weight of 0 This is done as the GPS part of the sample does not have an equivalent TSM group OSM non eth
84. utrition X Physical Activity X Smoking History X Disability X X see Health and Disability module Caring X X Partnership History new entrants X X of all ever in partnership Fertility History new entrants and X X has adopted or ever in partnership biological children Annual Event History interviewed in last X wave Current Employment X X Employees X X Self employment X X Commuting employee or self X Behaviour employed not in home Job Satisfaction employed X X Physical Work in paid work X Work Conditions employee X Non employment no paid work no job X X Second jobs X X Voluntary Work X Charitable Giving X Childcare Responsible for X X children Unearned Income X X and State Benefits Household Finances X X Personal Pensions X Savings X Retirement Planning age 45 50 55 60 X 65 and not retired Domestic Division of married civil X One item on hours Labour partnership of housework asked cohabiting and live in first 6 months at with partner wi Politics X X Political Engagement EMB GPC LDA X comparison General Election interviewed May X December 2010 Leisure Culture and X Sports Leisure Access X Positive and X Negative Events Interviewer X X 16 Observations Proxy asked about those X X not able to be interviewed in person In Wave 1 Language but see X Childhood language Migration History EMB GPC LDA X Employment Status History X asked of first 6 mont
85. vanced user to model Wave 1 household nonresponse taking into account the chance to be eligible among households of unknown eligibility To model eligibility we used predictors from the sampling frame and administrative neighbourhood data linked at a geographical level for detailed description see below After excluding ineligible addresses like businesses or demolished and nonexistent addresses the eligibility was modeled using only EMB households with known eligibility status either screened out or screened in This prediction was then extrapolated onto EMB households of unknown eligibility e g not contacted Given the limited number of selected addresses in Wales and Scotland and differences between countries in the available auxiliary variables see below we predicted eligibility using two models The first included common predictors for England and Wales and eligibility was predicted for these two countries The second was based on England Wales and Scotland using a more limited number of predictors Eligibility was predicted for Scotland only from this model Following this the probability of responding was estimated using backward stepwise logistic regression weighted by eligibility status where the ineligible were excluded those known to be eligible had a eligibility of one and those with unknown eligibility had a weight proportional to the predicted probability of being eligible obtained from the above model The predictors use
86. vel design weight corrects for e Unequal selection probability due to the boost in Northern Ireland The GPS selection probabilities in Northern Ireland are approximately twice those in other parts of the UK e Unequal selection probability due to the ethnic minority boost Selection probabilities in the EMB part of the sample vary considerably between areas depending on the estimated ethnic mix of the area and ethnic composition of the household Additionally households in high density areas with at least one ethnic minority member were weighted to account for combined probability of being selected as part of GPS or as part of EMB samples e The selection probability of households in a dwelling with more than 3 households or at an address with more than three dwellings is adjusted for the fact that only three such households were selected from the same address Individual level design weights correct for all the above with one specific difference non EM persons who live with EM persons in the same household have a chance to be selected only via the GPS part of the sample and not via EMB This means that non EM persons in the EMB who are TSMs are given a design weight of 0 while non EM persons in the GPS are given the household design weight The weights for EM persons are adjusted for their dual probability of being part of GPS and EMB Individual level design weights for those eligible to answer the Extra five minutes is similar to the a
87. veys will have the same main variable name though with a different wave prefix Since the last wave of BHPS was Wave 18 the wave prefix is R Thus if we wished to match Wave 2 work status b jbstat on the file B_INDRESP to previous wave values for the GPS sample we would match using pidp to A INDRESP and use the variable a jbstat while for the BHPS sample we would match using pid to RINDRESP and use the variable rjbstat 42 Variable information overview basic and derived variables Variable naming and labelling conventions Most variables have a mnemonic name Variables begin with a prefix designating the wave of data collection a_ for the first wave b for the second wave We have used to denote waves in general We have attempted to keep the names of variables that came from the BHPS the same for the convenience of analysts but this has not always been possible Analysts should consult the BHPS documentation https www iser essex ac uk bhps documentation volb index html Many derived variables are shown by the suffix _dv Derived variables include variables copied over from one file to another for analytic convenience variables that categorize a particular variable e g age category variables that combine information from multiple variables e g body mass index from self reported height and weight Information about how the derived variable is produced is shown in the notes for derived variables in th
88. viewing with the translated instruments A run through of all data collection instruments and procedures in 100 households called a dress rehearsal took place August September 2008 A pilot or run in for Wave 2 tested all instruments and data collection procedures For this wave the data collection also focused on assessing any problems with integrating members of the former BHPS sample component which includes a small segment conducted by telephone interviews In all 237 households were issued Of these 91 were households interviewed in the Wave 1 pilot The BHPS sample component was represented by households that were part of the BHPS between 1997 and 2001 the European Community Household Panel Households for which we had a telephone number were issued to telephone interview to test the telephone interview instruments and procedures The Wave 2 pilot took place September October 2009 Interviewers Because of the demanding nature of Understanding Society we tried to use interviewers of above average levels of experience and ability In Northern Ireland the majority of interviewers had worked on the Northern Ireland component of the BHPS the Northern Ireland Household Panel Survey and were familiar with the design and operation of Understanding Society In addition to general interviewer training interviewers working on Understanding Society attended a survey specific face to face briefing Generally around 12 20 interviewers attended e
89. y cases matched 46 merge m 1 a hidp using hhinfo drop this variable essential step drop merge save final1 replace clean up unwanted files erase hhinfo dta Example 2 Summarising individual level information at the household level In this example we will summarise individual level information within a household number of 18 24 year olds in the household and then match that onto the household level file use a hidp a hhsize using a hhresp ip clear sorta hidp save hhinfo replace use pidp a hidp a dvage using a indall ip clear create a variable that counts the number of 18 24year olds in each household bysort a egen n1824 sum a dvage 18 amp a lt 24 keep only first observation for every household bysort a keep if keep only household level information keep a hidp n1824 now merging this household information with the household level file sorta hidp merge 1 1 a hidp using hhinfo 47 drop merge save final2 replace erase hhinfo dta Example 3 Matching individuals within a household In this example we will match the information of wives onto that of their partners spouses Open the dataset with information on all persons in responding households and keep only those persons who have a spouse partner in the household use a hidp a pno a hgparta sexa dvage using a indall ip if a_hgpart gt 0 clear rename the prefix a to somethin
90. y sample only children born to an OSM mother will themselves become an OSM OSMs of all ages are followed for interview and remain eligible as long as they are resident within the UK They remain potentially eligible sample members for the life of survey The case may arise where the only OSM in the household is a child Other household members are then TSMs so long as they are co resident with the child and therefore eligible for interview even if the child is not yet old enough to be eligible for interview If the OSM child moves house they are followed to their new address and those living with the OSM child are eligible for interview If the OSM child moves into an institution where normally just the OSM PSM would be interviewed and not co residents a split off household is created containing only the OSM child and the household enumeration grid completed The child OSM is an eligible sample member even if they are not eligible for interview because of their age 21 Temporary Sample Members TSMs Any members of an enumerated household eligible for inclusion in the Ethnic Minority Boost sample at Wave 1 who are not from a qualifying ethnic minority are Temporary Sample Members TSMs at Wave 1 This was the only category of TSM at Wave 1 In all parts of the sample any new person found to be co resident in an OSM or PSM household after wave 1 is a TSM This would include any child born to an OSM father after wave 1 but not an OSM

Understanding Society - UKHLS: Wave 1-2, 2009

Contents

Download Pdf Manuals

Related Search

Related Contents