A user's guide
Contents
1. o sso o 012 2 Hoglibelthood GLE 9268 483 4059 of 4059 cages in use Add Iam Estimater Honlin The slope coefficient for standIrt for pupils from low intake ability schools is 0 455 For pupils from mid ability schools the slope is steeper 0 455 0 092 and for pupils from high ability schools the slope is steeper still 0 455 0 18 These two interaction terms have explained variabilty in the slope of standlrt in terms of a school level variable therefore the between school variability of the standlIrt slope has been substantially reduced from 0 015 to 0 011 Note that the previous contextual effects boysch girlsch mid and high all modified the intercept and therefore fitting these school level variables reduced the between school variabilty of the intercept We now have three different linear relationships between the output score normexam and the intake score standlIrt for pupils from low mid and high ability schools The prediction line for low ability schools is B cons B standlrt i The prediction line for the high ability schools is Bacons f standirt Byhigh B high standltt The difference between these two lines that is the effect of being in a high ability school is 10 B high B high standlrt We can create this prediction function by select the Model window Select Predictions Clear any existing prediction by clicking on variable Select Remove all explanatory variables from the menu tha2. A 65M 55 LI 7j Tot 65 Mi GG MM L3 12 j 17o 65 NI a7 fd 47 Lm 17 j tor 65 HI 1260M 126 L3 23 j 32o 65 Ni S0 a0 Lam 37 j 27 of 65 Mi So 3 LID 32 j Sof 65 Hi 42h 43 Lai 3 j Sf ot 66 Lim 3 j 30 6 A 520M 57 Lit B j 0 6 BH 102 M 102 L2 13 j 130f 65 Mi BS 64 L2I 18 16o 65 BH 1200M 120 Li 33 3o 65 BH 2810H 3 Lai 2 oi 65 BH 570M 57 Lai 33 Shor 65 Ai 770M 77 La 3 Shor 65 Lat 4 407 65 BH 73M 79 Lac 9 Sor 65 MA HM 34 L C 14 j 1407 65 EH 1380M 198 L20 13 j 190 BH SSM 55 LIE Mj Hoi bH OST OM 37 L2 3 ot AH FAM 79 L3 Mj Mol AN 280M 36 Lit 3 i Mat L30 5 Sof 65 Al 35 IM 35 LoD 10 1007 65 Al SO IM S50 LID 15 1500 RA SH gM 949 LD Wj Mot Wi SSM 35 Lao 5j Boat AA 730M 77 LiD j Moat AA DM a LID 3 j 35o 85 AA 30M 35 Late 40 j 40 oi HA TO ll 70 Mi 224 77 BH Said 54 AH S80 48 AO gM 71 The top summary grid shows in the total column that there are 4059 pupils in 65 schools The range column shows that there are maximum of 198 pupils in any school The details grid shows information on each school L2 ID means level 2 identifier value so that the first cell under details relates to school no 1 If when you come to analyse your own data the hierarchy that is reported does not conform to what you expect then the most likely reason is that your data are not sorted
3. the Stop button stops it and the More button resumes estimation after a stop The default method of estimation is iterative generalised least squares IGLS This is noted on the right of the Stop button and it is the method we shall use The Estimation control button is used to vary the method to specify convergence criteria and so on See the Help system for further details Click Start You will now see the progress gauges at the bottom of the screen R for random parameters and F for fixed parameters fill up with green as the estimation proceeds alternately for the random and fixed parts of the model In the present case this is completed at iteration 3 at which point the blue highlighted parameters in the Equations window change to green to indicate convergence Convergence is judged to have occurred when all the parameters between two iterations have changed by less than a given tolerance which is 10 by default but can be changed from the Options menu Click Estimates once more and you will see the parameter estimates displayed together with their standard errors as in the following screen the last line of the screen can be ignored for the time being 13 S Equations OF x normexam N XB Q normexam Jo cons 0 563 0 012 standlrt Boy 9 002 0 040 ug Egy ug NO ea NO Qo Q 0 566 0 013 p 2 0s 0 092 0 018 2 loglikelihood IGLS 9357 242 4059
4. we wish to specify a coefficient of standlrt which is random at level 2 To do this we need to inform MLwiN that the coefficient of Xj gt OF standirt should have the subscript j attached To do this Select the model menu Select the Equations window 21 Click Estimates until etc are displayed in black Click 2 Check the box labelled j school Click Done This produces the following result sf Equatane Vy NLIS oO Vy Fita Py iy Boy Bo by e oy Ay Ai tity f N O 9 Qy UET af Now that the model is becoming more complex we can begin to explain the general notation We have two explanatory variables x and x cons and standlrt Anything containing aQ subscript is associated with x and anything containing a 1 subscript is associated with x The letter u is used for random departures at level 2 in this case school The letter e is used for random departures at level 1 in this case student The parameters S and pare the fixed part regression coefficients associated with x and x They combine to give the average line across all students in all schools The terms u and ujare random departures or residuals at the school level from f and They allow the j th school s summary line to differ from the average line in both its slope and its intercept 22 The terms u and u follow a multivariate in this case bivariate Normal
5. 4 34 28 17 09 00 09 17 26 34 standirt Save your worksheet We will be using it in the next chapter What you should have learnt from this chapter e What is meant by contextual effects e How to set up multilevel models with interaction terms 13 Chapter 5 Modelling the variance as a function of explanatory variables From the fanning out pattern of the school summary lines seen in chapters 3 and 4 we know that schools are more variable for students with higher levels of standIrt Another way of saying this is that the between school variance is a function of standirt In MLwiN we always specify the random variation in terms of coefficients of explanatory variables the total variance at each level is thus a function of these explanatory variables These functions are displayed in the Variance function window Retreive the worksheet from the end of chapter 5 Select the Model menu Select Variance Function Click Name button in the Variance function window sf Yanance lunction 2 2 varle gons o geons The initial display in this window is of the level 1 variance In the present model we have simple constant variation at level 1 as the above equation shows Now In the level drop down list select 2 school of Venance lunciion E varin cons u standlrt oy gcois Je picons standlrt o standlrt cong standit vananga CUTp ho rons fi o SE of verience
6. by selecting highlight style 1 and clicking Apply If you do this you will see that the appropriate point in the top graph two lines in the middle graph and a set of points in the scatterplot have all become coloured red The individual school line is the the thinner of the two highlighted lines in the middle graph As would be expected from the fact that it has the highest intercept residual the school s line is at the top of the collection of school lines It is not necessary to highlight all references to school 53 To de highlight the school s contribution to the overall average line which is contained in dataset ds4 in the Customised graphs window Select dataset 3 Click on the other tab Click the Exclude from highlight box Click Apply In the caterpillar plot there is a residual around rank 30 which has very wide error bars Let us try to see why If you click on the point representing this school in the caterpillar plot the graph options window will identify it as school 48 Highlight the points belonging to this school in a different colour Using the graph options window in the in graphs box select highlight style 2 Click Apply The points in the scatterplot belonging to this school will be highlighted in cyan and inspection of the plot shows that there are only two of them This means that there is very little information regarding this school As a result the confidence limits for its residua
7. functions are often unreliable at extremes of the data to which they are fitted Another difficulty with using polynomals to model variances is that they may for some values of the explanatory variables predict a negative overall variance To overcome this we can use nonlinear negative exponential functions to model variance This is an advanced topic and for details see the Advanced Modelling Guide Yang et al 1999 We can construct a similar level 2 variance plot for the basic random slope model before extending the model by adding gender schgend and schav explanatory variables This can be illuminating because it shows us to what extent these variables are explaining between school differences across the range of standIrt This is left as an exercise for the reader but the graph comparing the between school variance for the two models is shown below st Graph display ertended a c T w oy D 2 i Al Of i if H standirt We see in both models schools are more variable for students with high standlrt scores The explanatory variables we added in the extended model explain about 25 of the between school variation across the spectrum of standIrt Complex variation at level 1 Until now we have assumed a constant variance at level 1 It may be that the student level departures around their school summary lines are not constant They may change in magnitude at different levels of standlrt or be larger fo
8. given the observed data and the estimated parameters of the model see Goldstein 1995 Appendix 2 2 In ordinary multiple regression we can estimate the residuals simply by subtracting the predictions for each individual from the observed values In multilevel models with residuals at each of several levels a more complex procedure is needed Suppose that yj is the observed value for the ith student in the jth school and that y is the predicted value from the average regression line Then the raw residual for this subject is 7 y y The raw residual for the jth school is the mean of these over the students in the school Write this as r Then the predicted level 2 residual for this school is obtained by multiplying r by a factor as follows O10 O uo O e0 n loj ee where n is the number of students in this school The multiplier in the above formula is always less than or equal to 1 so that the estimated residual is usually less in magnitude than the raw residual We say that the raw residual has been multiplied by a shrinkage factor and the estimated residual is sometimes called a shrunken residual The shrinkage factor will be noticeably less than 1 when 67 9 is large compared to 67 9 or when njis small or both In either case we have relatively little information about the school its students are very variable or few in number and the raw residual is pulled in towards zero In future residual
9. in the manner required by MLwiN In an n level model MLwiN requires your data to be sorted by level 1 within level 2 within level 3 level n There is a sort function available from the Data Manipulation menu We have now completed the specification phase for this simple model Itis a good idea to save the worksheet which contains the specification of the model so far giving it a different name so that you can return to this point in the manual at a later time Estimation We shall now get MLwiN to estimate the parameters of the model specified in the previous section We going to estimate the two parameters f and which in a single level model are the regression coefficients In multilevel modelling regression coefficients are referred 12 to as constituting the fixed part of the model We also estimate the variance of the school level random effects o and the variance of the pupil level random effects o2 The random effects and their variances are referred to as the random part of the model Click the Estimates button on the Equations window tool bar You should see highlighted in blue the parameters that are to be estimated Initially we will not estimate the 4059 individual pupil level random effects and 65 school level random effects we will return to these later The estimation process is iterative To begin the estimation we use the tool bar of the main MLwiN window The Start button starts estimation
10. order interaction exists and the view panel on the right of the window has been updated to include the option order 1 We just want to include main effects so In the View panel click on Main Effects The main effects and interactions window now displays a list of potential main effects Shain elects and mleractons At the moment no main effects are included to fit gender and school gender with boy and mixedsch as the reference categories click on the corresponding entries in the included column to produce the pattern T Main elects and mieradions The in higher column defines what categories are made available for higher order interactions This is useful when you have large numbers of categorical variables and the number of possible combinations for higher order interactions is very large To add the main effects to the model Click Build To view the model Select the Model menu Select Equations Click Names Which produces the following model sA E question normexam MXS cy homieraiy Sotas g atandi gurl ssboysch eirlsch Boy fa Fuy tey fy A thy H iy rs Nid a C Guo 1 Hi Guio Fut Mid i c F e i 9 Oo OQ c 2Haplizehhood IOLS 9357 1444059 of 4059 cases in use You can see that main effects for girl boysch and girlsch have been added Girl has subscript ij because it is a pupil level variable whereas the two school le
11. t Egy fy O S32 0 020 Huy uy 0 016 0 006 0 015 0 004 en NCO S Qe fo ssoco o12 2 Hogiibelihood TOLD 9278 443 4059 of 4059 cases in use foo die Mame O Aa Te Eaa Rw Gow Children attending mid and high ability schools score 0 067 and 0 174 points more than children attending low ability schools The effects are of borderline statistical significance This model assumes the contextual effects of school ability are the same across the intake ability spectrum because these contextual effects are modifying the intercept term That is the effect of being in a high ability school is the same for low ability and high ability pupils To relax this assumption we need to include the interaction between standlrt and the school ability contextual variables To do this Select the Main Effects and Interactions window Select Main Effects from the View panel Click the standIrt column for the mid and high entries Click Build The Main Effects and Interactions window should look like this E Main elects and The model converges to sA E question more xan BMD 0 HoOMMeXait fp cous g atandi 0 168 0 034 girl 0 184 0 098 boyeel H 161 0 O78 jgurlsch 14 4 0 094 md 0 290 0 106 High 0 00H 0 049 mid standlrt 0 180 0 055 thigh standlrt Boy 9 347 0 088 ug Eny Ary 9 455 0 042 H wy N o a 9 069 0 014 My 0 014 0 005 0 011 0 004 eny Nt 22
12. the plot type dropdown list select line Select the position tab h the grid click the cell in row 2 column 1 Click Apply This produces the following display si Graph display Now we can superimpose the overall average line by specifying a second dataset for the middle graph So that it will show up we can plot it in red and make it thicker than the other lines Select dataset ds4 at the left hand side of the Customised graphs window In the y dropdown list select c12 In the x dropdown list select standIrt In the plot type dropdown list select line Select the plot styles tab In the colour dropdown list select red In the line thickness dropdown list select 2 Select the position tab In the grid click the cell in row 2 column 1 I think the position should be OK following the previous manoeuvre Click Apply There is a lot more that MLwiN makes it possible to do with the graphs that we have produced To investigate some of this click in the top graph on the point corresponding to the largest of the level 2 residuals the one with rank 65 This brings up the following Graph options screen A Graph options Leme oul include highligh 1 Absorb into dumm highlightstie 2 Reset all ick on a point on a graph The box in the centre shows that we have selected the 53rd school out of the 65 whose identifier happens to be 53 We can highlight all the points in the display that belong to this school
13. the variability between different schools To do all this Click y either of the y symbols shown will do The Y variable dialogue box appears with two drop down lists one labelled y the other labelled N levels In the y list select normexam In the N levels list select 2 ij By convention the suffix i is used by MLwiN for level 1 and j for level 2 but suffixes can be changed as we will show later This reveals two further drop down lists level 2 j and level 1 i In the level 2 j list select school In the level J i list select student Click done In the Equations window the red y has changed to y in black indicating that the response and the number of levels have been defined Now we must define the explanatory variables Click x In the drop down list select cons Note that the fixed parameter box is checked by default each explanatory variable is assumed to have a fixed parameter We have just identified the explanatory variable X with a column of 1 s This vector of 1 s explicitly models the intercept Other software packages may do this for you automatically however in the interests of greater flexibility MLwiN does not Click Done The Equations window now looks like this sf Equals Y yf NAB y 0 Vi Pot o Eons Subs Nome Add Term Estimates Nonlinear We are gradually building equation 1 8
14. will mean shrunken residual Note that we can now estimate the level 1 residuals simply by the formula Coj i T Uoj L MLwiN is capable of calculating residuals at any level and of providing standard errors for them These can be used for comparing higher level units such as schools and for model checking and diagnosis Calculating residuals in MLwiN We can use the Residuals window in MLwiN to calculate residuals Let s take a look at the level 2 residuals in our model Select Model menu Select Residuals Select Settings tab sf Residuals raaiduals 1 eee fit SDicomperative ofresidualte siondesdisedidingnostic rasidualsia F nomal scores of residuals io a i nomal scores of sindardised residus F franks of residuals io F deletion residuals F levege values F biuence values jewel 1 student The comparative standard deviation SD of the residual is defined as the standard deviation of u and is used for making inferences about the unknown underlying value u given the estimate u The standardised residual is defined as SD uy and is used for diagnostic plotting to ascertain Normality etc As you will see this window permits the calculation of the residuals and of several functions of them We need level 2 residuals so at the bottom of the window From the level list select 2 school You also need to specify the columns into which the computed values of
15. with no random component Now run the model and view the estimates Press Start on the main toolbar Press Name then Estimates twice in the Equations window Which produces S Equations OF Xx normexam N XB Q normexam Jo cons 0 563 0 012 standlrt Boy 9 002 0 040 ug Egy ug NO ea NO Qo Q 0 566 0 013 4 0s 0 092 0 018 2 loglikelihood IGLS 9357 242 4059 of 4059 cases in use Fonts Subs Name Add Term Estimates Nonlinear Prep Clear O This should be familiar from the previous chapter The current model is a 2 level linear regression relationship of normexam on standirt with an average line defined by the two fixed coefficients By and B The model is made two level by allowing the line for the jth school to be raised or lowered from the average line by an amount uoj These departures from the average line are known as the level 2 residuals Their mean is zero and their estimated variance of 0 092 is shown in the Equations window With educational data of the kind we are analyzing they might be called the school effects In other datasets the level 2 residuals might be hospital household or area effects The true values of the level 2 residuals are unknown but we will often require to obtain estimates of them We might reasonably ask for the effect on student attainment of one particular school We can in fact predict the values of the residuals
16. 368 17 09 00 o9 17 standirt The lower curved line is the between school variation The higher straight line is the between student variation If we look at the equations screen we can see that 0 is zero to 3 decimal places The variance o acts as the quadratic coefficient in the level 1 variance function hence we have a straight line as the function is dominated by the other two terms The general picture is that the between school variation increases as standIrt increases whereas between student variation decreases with standIrt This means the intra school correlation school variance school variance student variance increases with standlrt Therefore the effect of school is relatively greater for students with higher intake achievements Notice as we pointed out earlier that for high enough levels of standIrt the level 1 variance will be negative In fact in the present data set such values of standIrt do not exist and the straight line is a reasonable approximation over the range of the data The student level variance functions are calculated from 4059 points that is the 4059 students in the data set The school level variance functions are calculated from only 65 points This means that there is sufficient data at the student level to support estimation of more complex variance functions than at the school level Lets experiment by allowing the student level variance to be a function of gender as well as standIrt We ca
17. Chapter 1 Random Intercept and Random Slope Models This chapter is a tutorial which will take you through the basic procedures for specifying a multilevel model in MLwiN estimating parameters making inferences and plotting results It provides both an introduction to the software and a practical introduction to multilevel modelling As we have seen multilevel models are useful in a very wide range of applications For illustration here we use an educational data set for which an MLwiN worksheet has already been prepared Usually at the beginning of an analysis you will have to create such a worksheet yourself either by entering the data directly or by reading a file or files prepared elsewhere Facilities for doing this are described at the end of this chapter The data in the worksheet we use have been selected from a very much larger data set of examination results from six inner London Education Authorities school boards A key aim of the original analysis was to establish whether some schools were more effective than others in promoting students learning and development taking account of variations in the characteristics of students when they started Secondary school The analysis then looked for factors associated with any school differences found Thus the focus was on an analysis of factors associated with examination performance after adjusting for student intake achievements As you explore MLwiN using the simplified data se
18. Click on Variable Select Include all explanatory variables Click on eo j to remove it In the output from prediction list select c11 Press calc This will form the predictions using the level 2 school residuals but not the level 1 student residuals For the overall average line we need to eliminate the level 2 residuals leaving only the fixed part of the model In the Predictions window click on ug to remove it In the output from prediction list select c12 Press calc Close the Predictions window The Customised graph window is currently showing the details of dataset ds2 the scatterplot With this dataset selected Click on the position tab In the grid click the cell in row 3 column 1 Press Apply The display now appears as follows st Graph display We have not yet specified any datasets for the middle graph so it is blank for the time being Here and elsewhere you may need to resize and re position the graph display window by pulling on its borders in the usual way Now let us plot the lines that we have calculated We need to plot c11 and c12 against standirt For the individual school lines we shall need to specify the group meaning that the 65 lines should be plotted separately In the Customised graphs window Select data set ds3 at the left of the window In the y dropdown list specify c11 In the x dropdown list specify standlrt In the group dropdown list select school In
19. In line 2 of the display the coefficient of standlrt has acquired a suffix j indicating that it varies from school to school In fact its mean from line 4 is 0 557 standard error 0 020 not far different from the model with a single slope However the individual school slopes vary about this mean with a variance estimated as 0 015 standard error 0 004 The intercepts of the individual school lines also differ Their mean is 0 012 standard error 0 040 and their variance is 0 090 standard error 0 018 In addition there is a positive covariance between intercepts and slopes estimated as 0 018 standard error 0 007 suggesting that schools with higher intercepts tend to some extent to have steeper slopes and this corresponds to a correlation between the intercept and slope across schools of 0 018 0 015 0 090 0 49 This will lead to a fanning out pattern when we plot the schools predicted lines As in the previous model the pupils individual scores vary around their schools lines by quantities e the level 1 residuals whose variance is estimated as 0 554 standard error 0 012 The quantity on the last line of the display 2 log likelihood can be used to make an overall comparison of this more complicated model with the previous one You will see that it has decreased from 9357 2 to 9316 9 a difference of 40 3 The new model involves two extra parameters the variance of the slope residuals u and their covariance with t
20. Press the Calc button in the prediction window The graph display window is automatically updated with the new values in column 11 to show the 65 parallel lines st Graph display 1 54 34 26 17 09 0 0 09 17 26 34 In this plot we have used the school level residuals u Residuals and their estimation are dealt with in more detail in the next chapter Student 7 in school j departs from the school j summary line by an amount Recalculate the predictions to include e as well as u as follows Click on 20 Press the Cale button sf Graph display 3 34 2 54 1 64 U 8 ani 0 4 HERA 1 64 A ETE 254 Wl vi 4 341 34 26 17 09 00 09 17 26 34 Which is a line plot through the original values of y i e we have predicted back onto the original data Experiment including different combinations of 8 i lo joy in the prediction equation Before pressing the calc button try and work out what pattern you expect to see in the graph window A Random slopes model The variance components model which we have just specified and estimated assumes that the only variation between schools is in their intercepts We should allow for the possibility that the school lines have different slopes as in Figure 1 3 This implies that the coefficient of standIrt will vary from school to school Still regarding the sample schools as a random sample from a population of schools
21. ained but they must be read in the right order and you are invited to save the current worksheet using a different name where necessary to preserve continuity Setting up a variance components multilevel model We now go through the process of specifying a two level variance components model for the examination data First close any open windows in the workspace Then Select Model menu Select Equations The following window appears st Egualiins v NCYB V Po AEA This window shows the nucleus of a model which you elaborate in stages to specify the one you want The tool bar for this window is at the bottom and we shall describe these buttons shortly The first line in the main body of the window specifies the default distributional assumption the response vector has a mean specified in matrix notation by the fixed part XB and a random part consisting of a set of random variables described by the covariance matrix Q This covariance matrix Q incorporates the separate covariance matrices of the random coefficientss at each level We shall see below how it is specified Note that y and x are shown in red This indicates that they have not yet been defined To define the response variable we have to specify its name and also that there are two levels The lowest level level 1 represents the variability between students at the same school the next higher level level 2 represents
22. d clicking This provides a means to edit data see the Help system for more details Having viewed your data you will typically wish to tabulate and plot selected variables and derive other summary statistics before proceeding to multilevel modelling Tabulation and other basic statistical operations are available on the basic statistics menu These operations are described in the help system In our first model we shall be looking at the relationship between the outcome attainment measure normexam and the intake ability measure standIrt and at how this relationship varies across schools The scatter plot of normexam against standlrt for the whole of the data looks like this a Graph display of x x lt v standlirt The plot shows as might be expected a positive correlation with pupils with higher intake scores tending to have higher outcome scores Our modelling will attempt to partition the overall variability shown here into a part which is attributable to schools and a part which is attributable to students We will demonstrate later in the chapter how to produce such graphs in MLwiN but first we focus on setting up a basic model You can now proceed straight away to the next section of this chapter or stop at this point and close MLwiN No data have been changed and you can continue with the next section after re opening the worksheet Tutorial ws Each of the remaining sections in this chapter is self cont
23. distribution with mean 0 and covariance matrix Q In this model we have two random variables at level 2 so Qis a 2 by 2 covariance matrix The elements of Q are var Uy om the variation across the schools summary lines in their intercepts var u o the variation across the schools summary lines in their slopes COV Uy O o the school level intercept slope covariance Students scores depart from their school s summary line by an amount ej We associate the level 1 variation with x because this corresponds to modelling constant or homogeneous variation of the student level departures This requirement can be relaxed as we shall see later To fit this new model we could click Start as before but it will probably be quicker to use the estimates we have already obtained as initial values for the iterative calculations Therefore Click More Convergence is achieved at iteration 7 In order to see the estimates Click Estimates twice if necessary Click Names To give 23 is Equations OF E4 A normexam N XB Q normexam jo cons 8 standlrt Boy 9 012 0 040 xg Egy By 0 557 0 020 u Hyl N 0 Q Q 9 090 0 018 ry 0 018 0 007 0 015 0 004 ea TNO Qe Q 0 554 0 012 Eons sabe Mone Aid Tom Estimates noninar cow __ You should compare this display with that for the model where we did not fit a random slope
24. h the predictions in column 11 against our predictor variable standIrt We can do this using the customised graph window Select the Graphs menu Select customised graph s This produces the following window 17 st Cusiomiced graph display 1 data set z Apply Layout Del data set Pre autosort on x Yes Details for for date set number ds 1 plotwhat plot style position erorbars ather none filter none 7 plot type point This general purpose graphing window has a great deal of functionality which is described in more detail both in the help system and in the next chapter of this guide For the moment we will confine ourselves to its more basic functions To plot out the data set of predicted values In the drop down list labeled y in the plot what tab select c11 In the neighbouring drop down list labeled x select standlrt In the drop down list labeled plot type select line In the drop down list labeled group select school This last action specifies that the plot will produce one line for each school For the present graph all the lines will coincide but we shall need this facility when we update our predictions to produce the school level summary lines To see the graph Click the Apply button The following graph will appear 18 sf Graph display Ud 26 17 09 00 08 17 26 3 We are now going to focus on the predictions window and
25. he fixed parameters this means that the majority of the schools do not differ significantly from the average line at the 5 level See Goldstein and Healy 1995 for further discussion on how to interpret and modify such plots when multiple comparisons among level 2 units are to be made Comparisons such as these especially of schools or hospitals raise difficult issues in many applications such as here there are large standard errors attached to the estimates Goldstein and Spiegelhalter 1996 discuss this and related issues in detail Note You may find that you sometimes need to resize graphs in MLwiWN to obtain a clear labeling of axes What you should have learnt from this chapter e Multilevel residuals are shrunken towards zero and shrinkage increases as nj decreases e How to calculate residuals in MLwiN Chapter 3 Graphical procedures for exploring the model Displaying graphs We have already produced a graphical display of the school level residuals in our random intercept model using the Residuals window to specify what we wanted MLwiN has very powerful graphical facilities and in this chapter we shall see how to obtain more sophisticated graphs using the Customised graphs window We will also use some of these graphical features to explore the random intercepts and random slopes models Graphical output in MLwiN can be described very appropriately at three levels At the highest level a display is essentially
26. he intercept residuals ug and the change which is also the change in deviance where the deviance for Normal models differs by a constant term for a fixed sample size can be regarded as a value with 2 degrees of freedom under the null hypothesis that the extra parameters have population values of zero As such it is very highly significant confirming the better fit of the more elaborate model to the data 24 Graphing predictions random slopes We can look at pattern of the schools summary lines by updating the predictions in the graph display window We need to form the prediction equation y Boj o Bi jy One way to do this is Select the Model menu Select Predictions In the predictions window click on the words Explanatory variables From the menu that appears choose Include all explanatory variables Click on e to remove it from the prediction equation In the output from prediction to drop down list select c11 Click Cale This will overwrite the previous predictions from the random intercepts model with the predictions from the random slopes model The graph window will be automatically updated If you do not have the graph window displayed then Select the Graphs menu Select customised graphs Click Apply The graph display window should look like this 25 a Graph display 17 26 34 The graph shows the fanning out pattern for the school prediction lines that is implied by the p
27. he standlrt column by clicking Press Build Note that we could have clicked on individual terms in Equations window and selected the delete term option However this would not have removed the terms from the main effects and interactions tables and every subsequent build would put them back into the model Contextual effects The variable schav is contructed by taking the average intake ability standIrt for each school based on these averages the bottom 25 of schools are coded I low the middle 50 coded 2 mid and the top 25 coded 3 high Let s include this categorical school level contextual variable in the model Select the Main Effects and Interactions window Select Setup from the View panel In the Categorical panel click on none Select schav from the list that appears Select Main Effects from the View panel Click the included column for the mid and high entries The main effects and interactions window should now look like this st himin ethects and mieracions i View cluded in higher slacdri setup order 1 onder Build M Chack for missing lower order components E Generis explanations Click Build Run the model by pressing more on the main toolbar sA E question OW ULE XA PHE oO nonMexan Sotos g atandi 0 167 0 034 purl 0 18 7 0 098 boyech 0 157 0 078 girlsch 0 067 0 085 mid 0 174 0 099 high Boy O 265 0 082 Hug
28. holds the data and other information in a series of columns These are initially named cl c2 but the columns can and should be given meaningful names to show what their contents relate to This has already been done in the Tutorial worksheet that you have loaded When a worksheetis loaded a summary of the variables shown below automatically appears si Names Mil E choal Sinbent Ihe ean cone simnl gender sched ait scha wriad cli ed ens cla c15 3 bt 1 2 934953 3 15952 1 1 1 a O 7S5900 0 637 6559 1 3 Each line in the body of the window summarises a column of data In the present case only the first 10 of the 400 columns of the worksheet contain data Each column contains 4059 items one item for each student represented in the data set There are no missing values and the minimum and maximum value in each column are shown Note the Help button on the tool bar The remaining items on the tool bar of this window are for attaching a name to a column We shall use these later You can view individual items in the data using the Data window as follows Select Data manipulation menu Select View or edit data si Daba 0 2513245 00 1340668 1 723862 0 957568 03da 1 734899 1039608 O 1290847 O 8995 TTF 1 219466 2 408692 0 6 107286 1 436687 A dnman i When this window is first opened it always shows the first three columns in the worksheet The exact number of items shown de
29. l are very wide and the residual itself will have been shrunk towards zero by an appreciable amount Next let us remove all the highlights from school 48 In the graph options window In the in graphs box select normal Click Apply Now let us look at the school at the other end of the caterpillar that with the lowest school level residual Click on its point in the caterpillar it turns out to be school 59 and in the Graph options window select highlight style 3 and click Apply The 8 highlighting will remain and the graphical display will look something like this having regard to the limitations of monochrome reproduction sE Graph display The caterpillar tells us simply that school 49 and 53 have different intercepts one is significantly below the average line the other significantly above it But the bottom graph suggests a more complicated situation At higher levels of standIrt the points for school 53 certainly appear to be consistently above those for school 49 But at the other end of the scale at the left of the graph there does not seem to be much difference between the schools The graph indeed suggests that the two schools have different slopes with school 53 the steeper To follow up this suggestion let us keep the graphical display while we extend our model to contain random slopes To do this From the Model menu select Equation Click on B and check the box labelled j school to make i
30. many data sets so in general there is no obvious text for the titles The existing titles appear because the graph was originally constructed by using the plots tab on the residuals window You can specify or alter titles by clicking on a graph In our case Click somewhere in the top graph to bring up the Graph options window Select the titles tab Edit the y title to be Intercept Edit the x title to be Slope Click Apply You can add titles to the other graphs in the same way if you wish Now the graphical display will look like this 11 st Graph display 12 The two schools at the opposite ends of the scale are still highlighted and the middle graph confirms that there is very little difference between them at the lower levels of standIrt School 53 stands out as exceptional in the top graph with a high intercept and much higher slope than the other schools For a more detailed comparison between schools 53 and 49 we can put 95 confidence bands around their regression lines To calculate the widths of the bands Select Predictions from the Model menu Edit the multiplier of S E to 1 96 From the S E of dropdown list select level 2 resid function From the output to dropdown list select column c13 Click Apply Now plot the bands In the Customised graphs window select dataset ds 2 Select the errors tab From the y error list se ect c13 From the y error list select c13 From the y err
31. n also remove the term which we have seen is negligible In the equations window click on 2 Check the box labeled i student The level 1 matrix Q is now a 3 by 3 matrix Click on the o term You will be asked if you want to remove the term from the model Click yes Do the same for 0 and O When you remove terms from a covariance matrix in the equations window they are replaced with zeros You can put back removed terms by clicking on the zeros Notice that the new level 1 parameter o is estimated as 0 054 You might be surprised at seeing a negative variance However remember at level that the random parameters cannot be interpreted separately instead they are elements in a function for the variance What is important is that the function does not go negative within the range of the data Note MLwiN by default will allow negative values for individual variance parameters at level 1 However at higher levels the default behaviour is to reset any negative variances and all associated covariances to zero These defaults can be over ridden in the Estmation Control window available by pressing the Estimation Control button on the main toolbar Now use the variance function window to display what function is being fitted to the student level variance vat e geons e standit e gt girl 2 2 ia ee 2 o cons 2 9 cons standlt c gul atandi From the equations
32. ns window that can be used to calculate predictions from the model and the Customised graphs window which is a general purpose window for building graphs that can be used to graph our predicted values Lets start by calculating the average predicted line produced from the fixed part intercept and slope coefficients f Select the Model menu Select Predictions Which produces 15 sf predidions variable fixed level 2 The elements of the model are arranged in two columns one for each explanatory variable Initially these columns are greyed out You build up a prediction equation in the top section of the window by selecting the elements you want from the lower section Clicking on the variable name at the head of a column selects all the elements in that column Clicking on an already selected element deselects it Select suitable elements to produce the desired equation Click on Click on Click on Names The prediction window should now look like this 16 fi y Bacons p Standhit vanlable cons standlrt fixed Bo Bi level 2 The only estimates used in this equation are By and B the fixed parameters no random quantities have been included We need to specify where the output from the prediction is to go and then execute the prediction In the output from prediction to drop down list select C11 Click Cale We now want to grap
33. nt level residuals Just as we can toggle between x s and actual variable names so we can show actual variable names as subscripts To do this Click the Subscripts button 10 Which produces aE guaina Pija E3 NOrMEX AM pudent sehor NCYB C NOTNEXAM enr En a Postudent school cons Dy standltt nent schol Bostudem schoot fo TE osho TF Eo P0 A 2 T oschoal i se oh E N O e Ostudent schoo gt i b5 e al a UU This display is somewhat verbose but a little more readable than the default subscript display You can switch between the subscript formats by pressing the subscripts button The screen shots in this chapter use the default subscript format You can gain more control over how subscripts are displayed by clicking on subscripts from the model menu sidant scrpaal Before running a model it is always a good idea to get MLwiN to display a summary of the hierarchical structure to make sure that the structure MLwiN is using is correct To do this Select the Model menu Select Hierarchy Viewer Which produces 11 si Heer anchy vierii ESMI B5 ESN L2 1 1 of 65 WI 73 el 73 L 6 Got 65 PI cO ie BO LI 11 j 11 65 N1 62 il G2 LI 16 j 160 65 Hi Sag 83 Li A j tot 65 M1 73 ie 73 LID Fh j Sot 65 M1 7S ie FS LID 3j sig amp HA 45 ll 3 LI 3j od 6 fo G5 156 05M 4055 LIE 2 j 201 65
34. number 14 Chapter 4 Contextual effects Many interesting questions in social science are of the form how are individuals effected by their social contexts For example e Do girls learn more effectively in a girls school or a mixed sex school e Do low ability pupils fare better when they are educated alongside higher ability pupils or worse In this section we will develop models to investigate these two questions Before we go on let s close all open windows by Select the Window menu Select close all windows Pupil gender and school gender effects We are now going use a new window which is useful for building models with categorical explanatory variables Select the Model menu Select Main Effects and Interactions The following window appears a Main effects and interactions Setup candidate variables Categorical Continuous none none I Check for missing lower order components IV Include vars in model This screen automates the process of creating sets of dummy variables and interactions between sets of dummy variables that are required for modelling categorical predictors To enter main effects for individual gender and school gender In the panel marked categorical click on none From the list that appears select gender Click on none again and select schgend Note that now that we have defined two cataegorical variables the possiblility for a 1
35. o tu Oj te Dij a MO i OU et e a N 0 Q Qe o Eome suss mame na Ton Eram Novice PH You may need to resize the window by dragging the lower border in order to see all the details or alternatively change the font size To replace y xo and x by their variable names Click the Name button normexam N XB Q normexam f cons standlrt Boy Bo Fuy tE gy E M N O Qu Qu ozo The Name button is a toggle clicking again brings back the x s and y s In summary the model that we have specified relates normexam to standirt The regression coefficients for the intercept and the slope of standIrt are 8 These coefficients define the average line across all students in all schools The model is made multilevel by allowing each school s summary line to depart be raised or lowered from the average line by an amount u The 7 th student in the j th school departs from its school s summary line by an amount e The information conveyed on the last two lines of the display is that the school level random departures u are distributed Normally with mean 0 and variance 67 9 and the student level random departures ex are distributed Normally with mean 0 and variance 07 9 the Q s can be ignored for the time being The uo one for each school are called the level 2 or school level residuals the eo one for each student are the level 1 or stude
36. of 4059 cases in use Fonts Subs Name Add Term Estimates Nonlinear Prep Clear O The first two lines of this display reproduce equations 1 8 with the actual names of the different variables filled in Recall that our model amounts to fitting a set of parallel straight lines to the results from the different schools The slopes of the lines are all the same and the fitted value of the common slope is 0 563 with a standard error of 0 012 clearly this is highly significant However the intercepts of the lines vary Their mean is 0 002 and this has a standard error in brackets of 0 040 Not surprisingly with Normalized data this is close to zero The intercepts for the different schools are the level 2 residuals uoj and these are distributed around their mean with a variance shown on line 4 of the display as 0 092 standard error 0 018 The variance appears to be significantly different from zero Judging significance for variances however and assignng confidence intervals is not as straightforward as for the fixed part parameters The simple comparison with the standard error and also the use of the interval and tests procedures see help system provides approximations that can act as rough guides We shall deal with this further when discussing the likelihood ratio statistic and also in the part of this guide which deals with simulation based techniques Of course the actual data points do not lie exactly on the straight line
37. on the row labeled 2 in the grid on the left hand side of the window Now use the y and x dropdown lists on the plot what tab to specify normexam and standIrt as the y and x variables in ds2 Next we need to specify that this graph is to separate from that containing the caterpillar plot To do this Click the position tab on the right hand side of the customised graph window The display can contain a 5x5 grid or trellis of different graphs The cross in the position grid indicates where the current data set in this case normexam standlIrt will be plotted The default position is row 1 column 1 We want the scatterplot to appear vertically below the caterpillar plot in row 2 column 1 of the trellis so Click the row 2 column 1 cell in the above grid Now to see what we have got Press the Apply button at the top of the Cutomised graph window and the following display will appear on the screen si Graph display OD L Oo O3 oo 0 a1 As a further illustration of the graphical facilities of MLwiN let us create a third graph to show the 65 individual regression lines of the different schools and the average line from which they depart in a random manner We can insert this between the two graphs that we already have First we need to calculate the points for plotting in the new graph For the individual lines Select the Model window Select Predictions
38. or boys the level 1 variance function has a negative slope indicating the boys who have high levels of standIrt are much less variable in their attainment We can graph these functions In the variance function window set output to list to c31 Press calc Select the customised graphs window Select the display used to plot the level 2 variance function Select the data set y c31 x standIrt In the group list select gender Click Apply 10 Which produces sf Graph display J a oO C m vie 3 4 26 1 7 09 0 0 09 standirt We see that the student level variance for boys drops from 0 8 to 0 4 across the spectrum of standlrt whereas the student level variance for girls remains fairly constant at around 0 53 We are now forming a general picture of the nature of the variability in our model at both the student and school levels of the hierarchy The variability in schools contributions to students progress is greater at extreme values of standIrt particularly positive values The variability in girls progress is fairly constant However the progress of low intake ability boys is very variable but this variability drops markedly as we move across the intake achievement range These complex patterns of variation give rise to intra school correlations that change as a function of standlrt and gender Modelling such intra unit correlations that change as a function of explanatory variables provides a u
39. or type list select lines Click Apply This draws 65 confidence bands around 65 school lines which is not a particularly readable graph However we can focus in on the two highlighted schools by drawing the rest in white Select the customised graphs window Select data set number 2 ds 2 From the colour list select white Click Apply 13 Graph display a oe T k E The confidence bands confirm that what appeared to be the top and bottom schools cannot be reliably separated at the lower end of the intake scale Looking at the intercepts and slopes may be able to shed light on interesting educational questions For example schools with high intercepts and low slopes plotting in the top left quadrant of the top graph are levelling up they are doing well by their students at all levels of initial ability Schools with high slopes are differentiating between levels of intake ability The highlighting and othere graphical features of MLwiN can be useful for exploring such features of complicated data See Yang et al 1999 for a further discussion of this educational issue What you should have learnt from this chapter e How to make different graphical representations of complex data e How to explore aspects of multilevel data using graphical facilities such as highlighting e With random slopes models differences between higher level units e g schools can not be expressed by a single
40. ositive intercept slope covariance at the school level To test your understanding try building different prediction equations in the predictions window before you press the cale button try and work out how the graph in the graph display window will change That concludes the second chapter It is a good idea to save your worksheet using the save option on the File menu What you should have learnt from this chapter You should understand e What a random intercept model is e What a random slope model is e The equations used to describe these models e How to construct estimate and interpret these models using the equations window in MLwiN 26 e How to carry out simple tests of significance e How to use the predictions window to calculate predictions from the model estimates 27 Chapter 2 Residuals In this chapter we will work through the random slope model again This time we shall explore the school and student random departures known as residuals Before we begin let s close any open windows Select the Window menu Select close all windows What are multilevel residuals In order to answer that question let s return to the random intercepts model You can retrieve one of the earlier saved worksheets or you can modify the random slopes model Select the model menu Select Equations Click on Uncheck the box labeled j school Click Done The slope coefficient is now fixed
41. output to nang The function shown is simply the variance of the sum of two random coefficients times their respective explanatory variables u cons and u standIrt written out explicitly Given that cons is a vector of ones we see that the between school variance is a quadratic function of standlrt with coefficients formed by the set of level 2 random parameters The intercept in the quadratic function is o2 the linear term is 26 and the quadratic term is o7 We can compute this function and the Variance function window provides us with a simple means of doing this The column in the window headed select cons standlrt and result are for computing individual values of the variance function Since standIrt is a continuous variable it will be useful to calculate the level 2 variance for every value of standIrt that occurs In the variance output to list on the tool bar select c30 Click Cale Now you can use the customised graph window to plot c30 against standIrt sf Graph display dy z i a a D 2 26 17 09 0 0 09 17 26 34 standirt The above graph has had the y axis rescaled to run between 0 and 0 3 The apparent pattern of greater variation between schools for students with extreme standIrt scores especially high ones is consistent with the plot of prediction lines for the schools we viewed earlier We need to be careful about over the interpretation of such plots Polynomial
42. pends on the space available on your screen You can view any selection of columns spreadsheet fashion as follows Click the View button Select columns to view Click OK You can select a block of adjacent columns either by pointing and dragging or by selecting the column at one end of the block and holding down Shift while you select the column at the other end You can add to an existing selection by holding down Ctrl while you select new columns or blocks The Font button which is present in several of the MLwiN windows can be used to make the characters in that window larger or smaller This can be useful when the space available for the windows is not too large The school and student columns contain identifiers normexam is the exam score obtained by each student at age 16 Normalised to have approximately a standard Normal distribution cons is a column of 1 s and standlIrt is the score for each student at age 11 on the London Reading Test standardised using z scores Normexam is going to be the y variable and cons and standIrt the x variables in our initial analysis The other data columns will be used in later sections of the manual Use the scroll bars of the Data window to move horizontally and vertically through the data and move or resize the window if you wish You can go straight to line 1035 for example by typing 1035 in the goto line box and you can highlight a particular cell by pointing an
43. r boys than girls In other words the student level variance may also be a function of explanatory variables Let s look and see if the pupil level variance changes as a function of standIrt To do this we need to make the coefficient of standIrt random at the student level To do this In the equations window click on Check the box labeled i student Which produces ft Eguatonsa normexam NYB 0 normerany Hogeo fF Standart feu fizboysch garlsch gud quad standiet Aphigh dich standirt Poy fo Thy Tey Dia i Fu F iy Hy Ni0 OL Guo wy Tuo Tu N O 0 Q ie a fy Sein Geil Joeli telikoa NALS 9263 2457 4059 of 4059 cases imuse Fonts Subs Mame Add Tara Nordine pup Now the coefficient of standlrt has a school level random term u and a student level random term eij attached to it As we have seen at the school level we can think of the variance of the u terms that is o 4 in two ways Firstly we can think of it as the between school variation in the slopes Secondly we can think of it as a coefficient in a quadratic function that describes how the between school variation changes with respect to standIrt Both conceptualisations are useful The situation at the student level is different It does not make sense to think of the variance of the e s that is o as the between student variation in the slopes This i
44. s because a student corresponds to only one data point and it is not possible to have a slope through one data point However the second conceptualisation where o is a coefficient in a function that describes how between student variation changes with respect to standIrt is both valid and useful This means that in models with complex level 1 variation we do not think of the estimated random parameters as separate variances and covariances but rather as elements in a function that describes how the level 1 variation changes with respect to explanatory variables The variance function window can be used to display the form of the function Run the model Select the variance function menu From the level drop down list select 1 student Which produces of Yenaace landiin ne 2 aana Valle gons e 8tandhit o gcons 2 2 26 5 cons standlt 5 Standlit select cons standit As with level 2 we have a quadratic form for the level 1 variation Let us evaluate the function for plotting In the output to drop down list select c31 Click calc Now let s add the level 1 variance function to the graph containing the level 2 variance function Select the customised graphs window Select the display used to plot the level 2 variance function Add another data set with y as c31 x as standIrt plotted as a red line Which produces si Graph display uE a Cc m Z i pi 44
45. s they vary about them with amounts given by the level 1 residuals eo and these have a variance estimated as 0 566 standard error 0 013 We shall see in the next chapter how MLwiN enables us to estimate and plot the residuals in order to obtain a better understanding of the model If we were to take children at random from the whole population their variance would be the sum of the level 2 and level 1 variances 0 092 0 566 0 658 The between school variance makes up a proportion 0 140 of this total variance This quantity is known as the intra school correlation It measures the extent to which the scores of children in the same school resemble each other as compared with those from children at different schools 14 The last line of the display contains a quantity known as twice the log likelihood This will prove to be useful in comparing alternative models for the data and carrying out significance tests It can be ignored for the time being This is another place where you would do well to save the worksheet Graphing Predictions Variance components We have now constructed and fitted a variance components model in which schools vary only in their intercepts It is a model of simple variation at level 2 which gives rise to the parallel lines illustrated in figure 1 2 To demonstrate how the model parameters we have just estimated combine to produce the parallel lines of figure 1 2 we now introduce two new windows the Predictio
46. s school We are currently using a reference group and three parameters to model a four entry table therefore because of the empty cells the model is saturated and no higher order intercations can be added The pupil gender and school gender effects modify the intercept standIrt 0 An interesting question is do these effects change across the intake spectrum To address this we need to extend the model to include the interaction of the continuous variable standIrt with our categorical variables To do this Select the Main Effects and Interactions window In the View panel select setup In the continous panel click on none From the list that appears select standIrt In the View panel select main effects The main effects screen now has a column for standlrt added Click on entries in the column to produce the following pattern st iain ethects and mieracions Check for missing lower oder components F Generale coplanar Click on Build The equations window will be automatically modified to include the thre new interaction terms Run the model press More on the main toolbar The deviance reduces by less than one unit From this we conclude there is no evidence of an interaction between the gender variables and intake score We can remove from the model by Select the main effects and interactions window Ensure main effects are selected in the view panel Deselect all entries marked with a X in t
47. seful framework when addressing interesting substantive questions Fitting models which allow complex patterns of variation at level 1 can produce interesting substantive insights Another advantage is that where there is very strong heterogeneity at level 1 failing to model it can lead to a serious model specification In some cases the mis specification can be so severe that the simpler model fails to converge but when the model is extended to allow for a complex level 1 variance structure convergence occurs Usually the effects of the mis specification are more subtle you can find that failure to model complex level 1 variation can lead to inflated 11 estimates of higher level variances that is between student heterogeneity becomes incorporated in between school variance parameters What you should have learnt from this chapter That variance functions are a useful interpretation for viewing variability at the different levels in our model How to construct and graph variance functions in MLwiN A more complex interpretation of intra unit correlation 12
48. t appears Click in turn on s Bs In the output from prediction to list select c30 Press Ctl N and rename C30 to predab Click Cale We can plot this function as follows Select the Customised Graph window Select display number 5 D5 In the y list select predtab In the x list select standIrt In the plot type list select line In the filter list select high Click Apply Which produces 11 st Graph display 0 4 i 34 2 6 1 7 19 00 09 1 7 Z6 34 standirt This graph shows how the effect of pupils being in a high ability school changes across the intake spectrum On average very able pupils being educated in a high ability school score 0 9 of a standard deviation higher in their outcome score than they would if they were educated in a low ability school Once a pupils intake score drops below 1 7 then they fare progressively better in a low ability school This finding has some educational interest but we do not pursue that here We can put a 95 confidence band around this line by Select the Predictions window Edit the multiplier S E of to 1 96 In the S E of list select Fixed In the corresponding output to list select c31 Click Cale Select the customised graph window Select error bars tab In the y errors list select c31 In the y errors list select c31 12 In the y error type list select lines Click Apply Which produces si Graph display 0 7
49. t random at level 2 Click Done Click More on the main toolbar and watch for convergence Close the Equations window Now we need to update the predictions in column c11 to take account of the new model From the Model menu select Predictions Click on uo and u to include them in the predictions In the Output from predictions dropdown list select c11 Click Cale Notice that the graphical display is automatically updated with the new contents of column c11 The caterpillar plot at the top of the display however is now out of date having been calculated from the previous model Recall we used the residuals window to create the caterpillar plot We now have two sets of level 2 residuals one giving the intercepts for the different schools and one the slopes To calculate and store these Select Residuals from the Model menu Select 2 School from the level dropdown list Edit the Start output at box to 310 Click Cale The intercept and slope residuals will be put into columns c310 and c311 To plot them 10 against each other In the Customised graphs window select dataset ds 1 and click Delete dataset From the y dropdown list select c310 From the x dropdown list select c311 Click Apply The axis titles in the top graph also need changing Note that if you use the customised graph window to create graphs no titles are automatically put on the graphs This is because a graph may contain
50. t you will also be imitating in a simplified way the procedures of the original analysis For a full account of that analysis see Goldstein et al 1993 Opening the worksheet and looking at the data When you start MLwiN the main window appears Immediately below the MLwiN title bar are the menu bar and below it the tool bar as shown File Edit Options Model Estimation Data Manipulation Basic Statistics Graphs Window Help IGLS Estimation control These menus are fully described in the online Help system This may be accessed either by clicking the Help button on the menu bar shown above or for context sensitive Help by clicking the Help button displayed in the window you are currently working with You should use this system freely The buttons on the tool bar relate to model estimation and control and we shall describe these in detail later Below the tool bar is a blank workspace into which you will open windows using the Window menu These windows form the rest of the graphical user interface which you use to specify tasks to MLwiN Below the workspace is the status bar which monitors the progress of the iterative estimation procedure Open the tutorial worksheet as follows Select File menu Select Open worksheet Select tutorial ws Click Open When this operation is complete the filename will appear in the title bar of the main window and the status bar will be initialised The MLwiN worksheet
51. the functions will be placed Click the Set columns button The nine boxes beneath this button are now filled in grey with column numbers running sequentially from C300 These columns are suitable for our purposes but you can change the starting column by editing the start output at box You can also change the multiplier to be applied to the standard deviations which by default will be stored in C301 Edit the SD multiplier to 1 96 Click Cale to calculate columns C300 to C308 Having calculated the school residuals we need to inspect them and MLwiN provides a variety of graphical displays for this purpose The most useful of these are available from the Residuals window by clicking on the Plots tab This brings up the following window One useful display plots the residuals in ascending order with their 95 confidence limit To obtain this click on the third option in the single frame residual 1 96 SD x rank then click Apply The following graph appears si Graph display This is sometimes known for obvious reasons as a caterpillar plot We have 65 level 2 residuals plotted one for each school in the data set Looking at the confidence intervals around them we can see a group of 10 or 15 schools at each end of the plot where the confidence intervals for their residuals do not overlap zero Remembering that these residuals represent school departures from the overall average line predicted by t
52. the graph display window Close the Equations Close the Customised graph window Arrange the predictions and graph display windows so that they are both visible If by mistake you click on the interior area of the graph display window a window offering advanced options will appear if this happens just close the advanced options window we will be dealing with this feature in the next chapter The line for the j th school departs from the above average prediction line by an amount u The school level residual u modifies the intercept term but the slope coefficient B is fixed Thus all the predicted lines for all 65 schools must be parallel To include the estimated school level intercept residuals in the prediction function Select the predictions window click on the term u The prediction equation in the top part of the predictions window changes from y B cons B standlrt to 19 y By cons B standlrt The crucial difference is that the estimate of the intercept B now has a j subscript This subscript indicates that instead of having a single intercept we have an intercept for each school which is formed by taking the fixed estimate and adding the estimated residual for school j Bo B tty We therefore have a regression equation for each school which when applied to the data produce 65 parallel lines To overwrite the previous prediction in column 11 with the parallel lines
53. vel variables have subscript j We can run the model and view the results by Click estimates until numbers appear in the equations window Press More on the main toolbar The model converges to the results below SAF quation more xan ME Oy HOMMeXAML fig cous 7 Standlrt 0 168 0 034 bsirly 1800 090 boyech 0 1750 O79 girlsch Boy O 189 0 031 g 29 fy O 554 0 020 Huy id i Ni o2 aN OFO0 0 0 G Wey CLOZ0 0 006 0 015 0 004 fen TNO Se Be o sso 0 012 1 K i 2 Hoplikelikhood NSLS 9281 120 4059 of 4059 cases in use 4 Ld Fonts Sube Meme Add Torm Estimates nomtiro Prit Chew The reference category is boys in a mixed school Girls in a mixed school do 0 168 of a standard deviation better than boys in a mixed school Girls in a girls school do 0 175 points better than girls in a mixed school and 0 175 0 168 points better than boys in a mixed school Boys in a boys school do 0 18 points better than boys in a mixed school Adding these three parameters produced a reduction in the deviance of 35 which under the null hypothesis of no effects follows a chi squred distribution with three degrees of freedom You can look this probability up using the Tail Areas option on the Basic Statistics menu The value is highly significant In the 2 by 3 table of gender by school gender there are two empty cells there are no boys in a girls school and no girls in a boy
54. what can be displayed on the computer screen at one time You can specify up to 10 different displays and switch between them as you require A display can consist of several graphs A graph is a frame with x and y axes showing lines points or bars and each display can show an array of up to 5x5 graphs A single graph can plot one or more datasets each one consisting of a set of x and y coordinates held in worksheet columns To see how this works Select the graphs menu Select customised graphs The following window appears sf Cusomi ed graph display 10 data sei Apply Layout Det data set Pile autosort on x dsa O W a Details for for data sei number ded 1 a EU cI05 L filter none plot type point This screen is currently showing the construction details for display D10 you may have noticed that the plot tab of the Residuals window in the previous chapter specified this in its bottom right hand corner The display so far contains a single graph and this in turn contains a single dataset dsl for which the y and x coordinates are in columns c300 and c305 respectively As you can check from the Residuals window these contain the level 2 residuals and their ranks Let us add a second graph to this display containing a scatterplot of normexam against standirt for the whole of the data First we need to specify this as a second dataset Select data set number 2 ds 2 by clicking
55. which assumes the simple level 2 variation shown in figure 1 2 We have specified the fixed parameter associated with the intercept and now require another explanatory variable Click the AddTerm button on the tool bar Click x Select standIrt Click Done The Equations window looks like this s8 Equals Y i HEYA y Vig Poo t Bit iy Hame Add Term Estimates Monlineer This completes the specification of the fixed part of the model Note that x has no other subscript but that x has collected subscripts ij MLwiN detects that cons is constant over the whole data set whereas the values of standIrt change at both level 1 and level 2 To define the random part Click p or x This redisplays the dialogue box for x seen earlier We wish to specify that the coefficient of x is random at both school and student levels Check the box labelled j SCHOOL Check the box labelled i STUDENT Click Done This produces si Egualins vy N B Q Vi Poto px 1ij Eonts Subs Heme Add Term Estimates Mourilinicas Prue We have now defined the model To see the composition of Bp Click the button on the tool bar You should now see the model as defined in equation 1 8 The and buttons control how much detail of the model is displayed Click a second time to reveal Jy N XB Q Yy Boy o t Pry Boy B
56. window we can see that 02 Oo O2 0 583 0 012 054 Substituting these values into the function shown in the variance function window we get the student level variance for the boys is 0 583 0 024 standIrt and for the girls is 0 583 0 054 0 024 standIrt Note that we can get the mathematically equivilent result fitting the model with the following terms at level 1 02 0 O This is left as an exercise for the reader The line describing the between student variation for girls is lower than the boys line by 0 054 It could be that the lines have different slopes We can see if this is the case by fitting a more complex model to the level 1 variance In the equations window In the level 1 covariance matrix click on the right hand 0 on the bottom line You will be asked if you want to add term girl standIrt Click Yes Run the model We obtain estimates for the level 1 parameters 02 0 9 O 15 0 584 0 032 0 031 0 058 The updated variance function window now looks like this si Vanance luncion soi 2 ama vat e 9 cons e standit e agit a pcons 26 5 cons standit 2c standht girl o atandi girl The level 1 variance for boys is now 0 584 2 0 032 standIrt 0 584 0 064 standlrt and for girls is 0 584 2 0 032 2 0 031 standlrt 0 058 0 526 0 02 standlrt We can see the level 1 variance for girls is fairly constant across standIrt F
Download Pdf Manuals
Related Search