COMPARE Compensation Analysis Software Manual
Contents
1. On this screen COMPARE requires analysts to acknowledge that given the complexity of multiple regression they are responsible for ensuring they understand the findings and that Biddle Consulting Group Inc and its staff will not be held liable for any use of these results Click on Agree to receive access to the full analysis report 14 IV Interpreting Compensation Analysis Results in COMPARE Once COMPARE completes all analyses and the user has agreed with the warning screen and caveats the following Excel worksheets will be presented These worksheets provide all of the information necessary to evaluate compensation disparities The default reports can be segmented into two major sections and will contain the following individual worksheets Section 1 Model Evaluation Section 2 Analysis Results Worksheet 1 Summary Worksheet 8 Stats No Exp Var Worksheet 2 Descriptives Worksheet 9 Stats Exp Var Worksheet 3 Correlations Worksheet 10 Comp Analysis Worksheet 4 Multicollinearity Worksheet 11 Liabilities Worksheet 5 Diagnostics Worksheet 6 Outliers Worksheet 7 Interaction As a matter of practice the analyst should review the worksheets in sequential order from left to right The reports in each worksheet are ordered to facilitate proper regression model evaluation prior to reviewing actual results of the analyses The information in each tab should be carefully reviewed and clearly understood before moving t2. Tests require that the data meet certain assumptions e g normality no extremely outliers COMPARE evaluates the data for any potential violations in assumptions and provides those results in the Diagnostics worksheet The components of this worksheet are presented below Diagnostics Report JobGroup Variable ValidN Non Missing Outliers Skewness Kurtosis 1 Salary 363 100 15 PPA 363 100 13 TIC 363 100 24 TJ 363 100 30 2 Salary 27 100 2 Kurt PPA 27 100 1 TIC 27 100 0 Skew TIJ 27 100 0 3 Salary 87 100 5 PPA 87 100 5 TIC 87 100 7 TIJ 87 100 5 The analyst should carefully review this report to ensure that data meets the necessary assumptions In particular note the following columns Valid N This column indicates the number of valid cases associated with each variable For example if a job title has 20 employees but the performance score is only available for 19 of them the number shown will be 19 rather than 20 Non Missing This column indicates the percent of the employees for whom valid data is available for all included variables In the example above it can be seen that 100 of the data was available for each employee for each explanatory factor in the three JobGroups Outliers 4 This column allows an analyst to identify the number of outliers that are contained in a particular SSEG This is important to note as outliers indicate employees who have very high or low va
3. Simply create a variable or variables to identify SSEGs e g job title region business unit AAPs The SSEG variable can be Character String lt STR gt Numbers lt NUM gt or Binary lt BIN gt Employee Identifier This is an optional specification but it is helpful to include employee ID in the final reports e g identifying outliers liabilities distribution This field must be Numeric lt NUM gt ID 41 78 48 22 81 90 92 84 43 Salary 34 641 34 107 34 550 35 832 34 884 35 309 34 016 34 286 35 809 Tenure Performance Gender 1 Man 1 Woman 4 Man 5 Man 5 Woman 5 Man 2 Woman 2 Woman 5 Woman GenderCoded Step 2 Opening the Data to be Analyzed COMPARE requires that the data to be analyzed be opened prior to starting the program Ifthe data to be analyzed is opened then it will be available to be selected in the Select Workbook drop down discussed in Step 4 below Step 3 Opening the COMPARE Excel Workbook COMPARE was developed to be seamlessly integrated with Excel To access COMPARE s suite of statistical modules simply open it as any Excel workbook and the following screen should appear Click Run to begin Copyright 2009 COMPARE o Compensation Analysis Reporting Software Exit Run Biddle Consulting Group Inc www biddle com Step 4 Configuring the COMPARE Data Setup The first step is to specify
4. below 2 Open the Data to be Analyzed Before starting COMPARE the prepared data file must be opened 3 Open the COMPARE Excel Workbook 4 COMPARE Data Setup Specify e Data Workbook e Data Worksheet e Output Workbook 5 COMPARE Analysis Setup Specify the compensation model including e Salary Required e Explanatory Variable s Optional e Comparison Group Variable Required e Analyze By Variable s Optional e Employee Identifier Optional 6 COMPARE Report Setup Optional Customize the reports to be generated and some details in each of the reports if desired 7 COMPARE Analysis Specifications Optional Customize analysis specifications if desired 8 Run the Analysis COMPARE begins analyzing the data and generating reports Step 1 Preparing the Data The purpose of data preparation is to ensure that the data to be analyzed conforms to COMPARE s required data specifications 1 Data Worksheet Setup To ensure that COMPARE can import the data the data worksheet must conform to the following basic rules e Data Rows Variable Names must be on Row 1 e Data Columns Empty columns are not allowed COMPARE will ignore all data columns to the right of any empty columns e Naming Variables No spaces are allowed in Variable Names For example GenderCoded without spacing 1s acceptable but Gender Coded with spacing is not Data and Variable Setup Compensation analyses have very specifi
5. 4 Men y O ji O Women er aap Bo E un Linear Men Z near Women IL 45 000 40 000 O l 3 4 5 9 Performance Appraisal Score Note Scatterplots are not currently included within COMPARE In this example COMPARE found the gender group by Performance Appraisal Score interaction to be significant Salary increased with higher Performance Appraisal Scores for men but remained constant across the range of Performance Appraisal Scores for women Clearly Performance Appraisal Scores operate differently for men than women when it comes to making pay predictions Important Note This is a minor treatment of a very complex topic In general whenever an interaction is present please consult professional statisticians who have experience with such matters Interactions can be problematic on three levels First it may be an indicator of a tainted variable Second this could be interpreted by some as prima facie evidence of discrimination Third it is extremely difficult to identify the dollar amounts needed to rectify the situation and to whom those dollars should be allocated The mechanics of testing for interactions is fairly straightforward but proper interpretation and understanding of interactions requires a firm grasp of multivariate statistics Consequently it is impossible to fully test an interaction and provide simple palatable interpretation This is full disclosure and warning for the beginning to mod
6. Yes 1 348 823 71 Total Liabilities Regression Focal No 0 00 Total Liabilities t Test Reference Yes 89 649 33 Total Liabilities Regression Reference Yes 11 027 23 Liability Excluded N 10 e Comp Analysis This section of the report provides several high level perspectives on compensation differences o Signif t Test Provides a count and percentage of SSEGs e g job groups job titles etc that exhibited significant pay disparities based upon t Test results o Workforce in Groups with Signif t Test Difference Provides a count and percentage of the total workforce in SSEGs with significant pay disparities based upon t Test results o Signif Regression Provides a count and percentage of the SSEGs that exhibited significant pay disparities based upon multiple regression results Workforce in Groups with Signif Regression Difference Provides a count and percentage of the total workforce in SSEGs with significant pay disparities based upon multiple regression results Non Signif t Test but Signif Regression Provides a count and percentage of the number of SSEGs that were not significant based upon Test results but became significant when using multiple regression Workforce in Groups with Non Signif 1 Test but Signif Regression Difference Provides a count and percentage of the workforce in SSEGs where pay disparities were not significant when using a t Test but became significant when using multiple re
7. gender race Those factors that share a statistically significant relationship at p lt 0 05 are identified with a single asterisk and those that have a statistically significant relationship at the p lt 0 01 level are identified with a double asterisk Next examine the correlations between each of the other factors If a correlation is very high i e approaching 1 00 it indicates a potential problem with collinearity which would mean that both variables cannot be included in the model without leading to serious problems with the regression model COMPARE includes specific measures of multicollinearity in a separate tab but this is a useful visual tool as well If two explanatory factors appear to be too highly related e g r gt 80 it may be best to remove the variable of the two with the weakest relationship to compensation 21 Worksheet 4 Multicollinearity The Multicollinearity worksheet helps to identify signs of excessive relationships between explanatory variables The components of this worksheet are presented below Multicollinearity Report Varl VI Var V2 372 es a NE aa i PPA TIC 0 100 0 849 0 851 1 116 PPA TU 0 042 0 849 0 854 TIC PPA 0 100 0 217 0 250 5 978 TIC TU 0 902 0 217 0 373 TH PPA 0 042 0 062 0 184 5 929 TU TIC 0 902 0 062 0 317 The following guidelines are adopted by COMPARE 22 Variance Inflation Factor VIF This is a measure of the imp
8. on counts of Focal and Reference members and percent difference in compensation High potential liability areas are shaded Worksheet 3 Correlations The Correlations worksheet is particularly useful for helping identify those factors that are most powerful in explaining the company s pay practices The components of this worksheet are presented below Correlation Report Variable Name Salary PPA TIC TI Minority Salary 0 849 0 217 0 062 0 034 PPA 0 849 0 1 0 042 0 091 TIC 0 217 0 1 0 902 0 813 TIJ 0 062 0 042 0 902 0 856 Minority 0 034 0 091 0 813 0 856 Notes Significant at the 05 level Significant at the 01 level The correlation coefficients included in the Correlations worksheet allow the analyst to determine whether e Explanatory variables are related to compensation e The focal variable i e gender minority status etc is related to compensation or e There are relationships between the various predictors that could lead to multicollinearity When reviewing the Correlation Report pay particular attention to the second column to the immediate right of the Variable List column This column indicates the correlations between each factor and the variable being used to measure employee compensation Ideally those variables that are most highly correlated with compensation are the explanatory i e job related variables and not the protected characteristic e g
9. Compare 2 0 Compensation Analysis Software Technical User Manual Version 11 Biddle Consulting Group Inc Copyright 2011 Biddle Consulting Group Inc Biddle Consulting Group Inc 193 Blue Ravine Road Suites 250 270 Folsom Ca 95630 1 800 999 0438 All rights reserved No part of this manual may be reproduced stored in a retrieval system or transmitted in any form or by any means electronic mechanical photocopying recording or otherwise without the prior permission of the publisher Table of Contents I Conducting and Interpreting Analyses in COMPARE 1 Introduction cia 1 II Installing COMPARE ooo 3 Minimum System Requirements occccccccccccccccncncncncnoncncncncncncnenonenenenenonos 3 111 Steps for Conducting Compensation Analyses in COMPARE 5 Step 1 Preparig the Dita ii oia 5 Step 2 Opening the Data to be Analyzed ooooocooccocccoccnocanonnnonnnnanonanonanconnconn ccoo en 7 Step 3 Opening the COMPARE Excel WorkbooKk oooooooiocoioocnooccooncoonoconoconocononos 7 Step 4 Configuring the COMPARE Data Setup ooooocccoccoocononononnonn nono nonnnononocnonononos 8 Step 5 Configuring the COMPARE Analysis Setup cooooooonionnconnconnconncon nono noconocononos 9 Step 6 Configuring the COMPARE Report Setup ooooooccooncconoconoconoconaconaconnnnanonoos 11 Step 7 Configuring the COMPARE Analysis Specifications uee 12 Step 8 Running the Analy
10. PARE Introduction This manual serves as an introduction and step by step guide to conducting compensation analyses using BCG s COMPARE compensation analysis software package It is strongly recommended that the information presented throughout this book be mastered prior to conducting an analysis with COMPARE The nature of statistical software is that it can make complex analyses appear simple However given the data assumptions and analytical requirements discussed previously much can go wrong in the course of a multiple regression analysis Relevant issues must be identified and addressed to ensure that results are as accurate valid and defensible as possible WARNING Multiple Regression is an extremely powerful and highly complex statistical procedure While it is the ideal tool for determining whether differences in pay are due to legitimate factors or possibly due to protected employee characteristics if it is not applied correctly it can lead to misleading and or unsupportable results If the analyst is not completely knowledgeable and experienced with the concepts presented in this manual or if the employer is considering making changes to employee compensation as a result of the analyses it is strongly encouraged that assistance be enlisted from an attorney or a statistical professional familiar with the use of multiple regression in compensation analysis II Installing COMPARE COMPARE was developed to seamles
11. Pay Disparity 1 Category Identifies the Multicollinearity High i a gt 0 8 A lo gt category of analysis Diagnostics No Missing Data Yes 107 Outliers Outliers performed Yes 1 Skew Skewness Yes 1 Kurtosis Kurtosis 2 Problem Identifies 3 3 100 Groups Analyzed with t test 477 477 100 Workforce Analyzed with t test whether a prob lem 3 3 100 Groups Analyzed with Regression s 477 477 100 Workforce Analyzed with Regression exists by analysis Interaction Yes 1 3 33 Potentially Tainted Variable category Results Overview 3 Severity Identifies the CATEGORY PROBLEM SEVERITY DESCRIPTION i Comp Analysis Yes 3 3 100 Signif t test severity of the problem Yes 477 477 100 Workforce in groups with Signif t test Yes 2 3 67 Signif Regression by analysis category Yes 390 477 82 Workforce in groups with Signif Regression BR E No 0 3 0 Impacted Groups t test nonsig Regression signif 4 Description Provides a No 0 477 0 Impacted Persons t test nonsig regression signif z une Yes 1 3 33 Impacted Groups t test signif Regression nonsig brief description of the Yes 87 477 18 Impacted Persons t test signif Regression nonsig Liabilities Yes 2694480 55 Total Liabilities t test Focal problem and some Yes 1348823 71 Total Liabilities Regression Focal ioh 1 No 0 Total Liabilities t test Reference high level details Yes 89649 33 Total Liabilities Regression Reference The summary and all
12. These differences are based on descriptive statistics but are very helpful at identifying potential problem areas e Correlations YES indicates a statistically significant correlation between the comparison group membership e g whites or minorities and compensation A significant relationship between these two variables is a potential indicator of discrimination e Multicollinearity YES indicates the existence of multicollinearity and the model s findings may be suspect e Diagnostics YES indicates a problem with the indicated diagnostic as well as the number of employee groups which were analyzed with either a Test or multiple regression e Interaction YES indicates one or more variables were interacting with the comparison group and may be potentially tainted 16 Results Summary Category Problem Severity Description Comp Analysis Yes 3 0f3 100 Signif t Test Yes 477 477 100 Workforce in Groups with Signif Test Difference Yes 2 3 67 Signif Regression Yes 390 477 82 Workforce in Groups with Signif Regression Difference No 0 3 0 Non Signif Test but Signif Regression No 0 477 0 Workforce in Groups with Non Signif Test but Signif Regression Difference Yes 1 3 33 Signif t Test but Non Signif Regression Yes 87 477 18 Workforce in Groups with Non Signif Regression but Signif t Test Difference Liabilities Yes 2 694 480 55 Total Liabilities t Test Focal
13. VII enforcement agencies pursue and successfully bring claims where the impact has been against reference group members Consequently COMPARE computes liabilities for both Focal and Reference group members by default The analyst who wishes only to compute liabilities for Focal group members may choose to uncheck this selection e Liabilities Distribution There are three methods of distributing liabilities o Evenly Distributed All Individuals Evenly distributes liabilities to all individuals in the negatively impacted group o Evenly Distributed Below Average COMPARE identifies individuals in the impacted group who are below the group average and evenly distributes the liability to them o Proportionally Weighted by Impact COMPARE identifies all individuals in the negatively impacted group and computes the difference between their current salary and their predicted salary based on a model using only job related qualifications and not gender minority status Only individuals in the negatively impacted group who fall below their predicted compensation are identified as eligible to receive a portion of calculated liability amount Among those individuals liability amounts are computed in proportion to the difference between the individuals current salary and their predicted salary see Compensation Analyses A Practitioner s Guide to Identifying and Addressing Compensation Disparities by Biddle Consulting Group for a full de
14. act of collinearity among the variables in a regression model It is always a number that is equal to or greater than 1 00 Larger numbers indicate greater impact of collinearity COMPARE uses a default criterion of 10 SPSS provides conservative guidelines that Condition Index values over 15 indicate possible problems with collinearity while values over 30 suggest a serious problem with collinearity Correlation Between Factors A second diagnostic of collinearity is based on the size of the correlation between factors If the correlation is greater than or equal to 0 80 it is a sign that the two variables are too highly related to both be included in the model In these situations it is recommended that the variable with the strongest relationship to compensation be retained Multicollinearity Determination If VIF is greater than 10 0 or the correlation is greater than or equal to 0 80 multicollinearity is likely to exist if both of the factors remain in the model Important Note If multicollinearity exists in a model it means that there is an issue with the data that will result in problems and in faulty results If multicollinearity exists it must be evaluated before trusting the results of any analysis If the analyst does not fully understand the issues and methods for addressing multicollinearity it is strongly recommended that a professional compensation statistician be consulted Worksheet 5 Diagnostics Regression and
15. ation However if Job Group is selected first followed by Location then the results will display all Locations within each Job Group Employee Identifier Optional Unique employee IDs are helpful references when interpreting the Outliers and Liability computation reports If an Employee Identifier is not specified COMPARE will generate a unique and random Employee ID for each record To specify an employee identifier select the variable from the Available Variables box and click the right arrow button next to the Employee Identifier box Note that only the Numeric lt NUM gt datatype is permissible in this specification Analyses By default COMPARE will run a comprehensive set of model diagnostics and statistical tests for analyzing pay disparities including Descriptives Diagnostics Correlations Interactions Multicollinearity and Regression Of the six analyses Descriptives and Regression are required for each COMPARE analysis The analyst may choose not to run the remaining four analyses Setup Advanced User Specifications COMPARE has a very efficient user interface requiring minimal user specifications It should be recognized however that highly complex statistical procedures are run from this relatively simple interface Analysts may choose to rely on default settings or specify additional analyses to better suit their statistical needs The default advanced user specifications have been set to BCG best practice stan
16. ble 1 272 34 787 6 0 8 0 7 0 0 0 00 5 856 54 218 41 500 7 0 4 0 6 0 0 0 00 5 840 02 72 30 415 5 0 3 0 4 0 0 0 00 5 753 71 299 16 130 2 0 11 0 8 0 0 4 194 73 5 724 33 279 16 113 2 0 7 0 8 0 0 4 225 94 5 702 30 207 16 002 2 0 6 0 7 0 0 4 429 77 5 658 23 204 16 048 2 0 9 0 7 0 0 4 345 30 5 638 03 393 16 073 2 0 10 0 7 0 0 4 299 39 5 601 30 441 16 389 2 0 9 0 8 0 0 3 719 14 5 531 53 55 16 512 2 0 11 0 8 0 0 3 493 28 5 529 69 355 16 573 2 0 9 0 8 0 0 3 381 26 5 515 00 150 16 638 2 0 11 0 8 0 0 3 261 91 5 461 75 201 16 113 2 0 5 0 6 0 0 4 225 94 5 452 57 1 t Liable Provides computed liability by employee based on t Test model 2 R Liable Provides computed liability by employee based on regression model If Small N is listed in the results it indicates that the number of employees in the group was too small to make the necessary calculations Values are only listed when an employee s pay is sufficiently below their expected salary to warrant further consideration for a pay adjustment based upon the analysts chosen level of allowable disparity The other columns in this report are presented to assist the analyst in interpreting the liabilities should a cohort analysis be necessary Note that the R liabilities are sorted in descending order Ifa warning message appears on the header of this report stating Warning X cases were excluded from liability computations Potential exc
17. c requirements and very specific data types for each field included in the analysis For example the salary variable can only be numeric e g 50 000 and the comparison group variable which identifies the group membership of each record must be binary e g 0 Women 1 Men The following section will begin with a discussion of datatype then move into the specifics of COMPARE s data requirements Datatype COMPARE was designed to import three datatypes as shown below Datatype Details Character Strings lt STR gt Alphabetical characters and words e g Unit01 Job2 Numeric lt NUM gt Numbers only e g 50 000 3 17 Binary lt BIN gt Only 0 and 1 Data Requirements COMPARE only allows certain datatypes for each of the fields on the Analysis Setup screen Step 5 The following table is a quick reference of the variable fields on Analysis Setup and the permissible datatype s Datatype Variable Fields Character String Numeric Binary Required lt STR gt lt NUM gt lt BIN gt 1 Salary Y Y 2 Explanatory Y Y 3 Comparison Group Y Y 4 Analyze By Y Y Y 5 Employee Identifier Y 1 Salary The salary variable is a required field and must be numeric As a word of caution some databases may include characters in the salary variable to track events and COMPARE will treat that variable as a character strings lt STR gt Consequently the analyst will not be able to specify that variable into the salary fi
18. ce This essentially means that employees are grouped so that they are performing essentially the same duties have the same minimum qualifications require the same levels of supervision and have the same levels of responsibility This is not a statistical test it only provides a high level indicator for potential disparity and should only be used as a guide COMPARE tests for interactions at the aggregated level across all SSEGs Although COMPARE is equipped to evaluate interactions for each SSEG the sample size requirements to test for interactions require a more conservative implementation At the aggregated level COMPARE is most capable of testing and identifying potential interactions 33 34
19. dards and are suitable for most compensation analyses Advanced users may adjust these settings by clicking Setup to access two power user specification screens 1 Report Setup and 2 Analysis Specifications Each of these will be detailed in the next section 10 Step 6 Configuring the COMPARE Report Setup The Report Setup screen allows the analyst to customize the output reports There are five major specifications that the analyst can customize Output Reports Outlier Analysis Group Heading Specifications Data Worksheets and Report Header Report Setup Compare 2 x Biddle Consulting Group Inc Copyright 2009 1 Qutput Reports By default Report Setup Analysis Specifications COMPARE generates a Output Reports Outliers Analysis comprehensive library of reports However there are three reports that are not automatically generated These IV Descriptive T Full Regression l Full Liability I Outlier z Scores M Summary l Full t test gt r Group Heading Specifications r Data Worksheets are all raw statistical outputs Focal Group 1 1 e 9 Female Minority 40 diia Analysis Data from the regression analysis t FERRE Test analysis and the liabilities Reference Group reference l Copy of Original 2 e g Male Non Minority lt 40 panna analysis r Report Header Analysts who wish to evaluate Privileged amp Confidential Attorney Client Work Product t
20. dentifier Optional 6 Employee Identifier gt i Analyses To specify a variable field IV IV Diagnostics IV Correllations simply click the arrow IV Interactions Y MultiCollinearity IV buttons associated with each of the variable fields To add a variable to a lt Previous field click the right arrow Help button gt To remove a variable from a field select the variable and click the left arrow button E The six fields are discussed in detail below 1 Available Variables The box on the left side of the screen presents a list of the fields that were successfully imported into COMPARE These variables are available for the compensation analysis 2 Salary Required From the Available Variables box select the salary compensation variable by clicking on the variable and click the right arrow button gt next to the Salary box Note that only the Numeric lt NUM gt datatype is permissible 3 Explanatory Variables Optional If there are variable s that account for potential pay differences the analyst may specify them in the Explanatory Variables box For each explanatory variable select it from the Available Variables box and click the right arrow button gt next to the Explanatory Variables box Note that only the Numeric lt NUM gt datatype is permissible The following explanatory variables will be used in examples throughout this manual e TIC Time in Company e TIJ Time in Job e PPA P
21. egatively impacts Focal members 3 815 73 and after controlling for the explanatory variables the pay disparity is even greater 5 493 63 JobGroup 2 The 1 Test indicates the pay disparity negatively impacts Focal members but after controlling for the explanatory variables the pay disparity is actually negatively impacting Reference members This is not as uncommon as one would believe JobGroup 3 The t Test indicates the pay disparity significantly negatively impacts Focal members but after controlling for explanatory variables the pay disparity is no longer significant p 0 21 Each column of the report is discussed in detail below 1 2 3 Columns left ofthe Data column are SSEG identifiers Data Indicates the statistics that are reported in the Focal Reference and Grand Total columns Focal Reference Grand Total These are the default headers Analysts who customize the Report Setup can expect to see the headers that they specify t Impact Difference Provides results of the Test analyses including the negatively impacted group and the raw mean difference between the two groups Reg Impact Difference Provides results drawn from the regression analyses including the negatively impacted group and the adjusted mean difference between the two groups after controlling for the explanatory variables t Test p value Provides the p value for the Test i e whether the model is a significant predictor
22. eld 2 Explanatory Explanatory variables help to explain potential pay differences e g tenure performance years of education though they are optional COMPARE will only allow Numeric lt NUM gt and Binary lt BIN gt datatypes to be specified in this field If the analyst wants to include explanatory variables in the compensation model they must ensure they are Numeric or Binary datatypes 3 Comparison Group COMPARE tests for pay disparities between two groups Focal and Reference e g men women The comparison group variable identifies the membership of each data record into one of two groups therefore it can only be a Binary lt BIN gt datatype As part of data preparation the analyst must code out members of each group as follows e Focal members e g women minority individual minority group 40 0 e Reference members e g men white lt 40 1 4 A sample dataset may look like the following Sample COMPARE Dataset Analyze By If employees can be separated into Similarly Situated Employee Groupings SSEGs it is important to run compensation analyses for each of these SSEGs Appropriate SSEGs are critical to a valid compensation analysis Analysis results are invalid and misleading if they are drawn from data with employees who do not perform similar work functions or who are expected to have similar pay such as if office assistants research directors and interns were all clustered into one analysis
23. erately advanced analysts Proper evaluation of interactions and their impact on employee compensation is critical to proper compensation analysis 27 Section 2 Evaluating the Statistical Results Worksheet 8 Stats No Exp Var The raw 1 Test statistics are presented in the Stats No Exp Var No Explanatory Variables worksheet Because Tests are simply unique regression models with no explanatory variables the results are presented in a general regression framework which provides more diagnostic information Proper interpretation of the report requires a firm understanding of regression statistics The report presents the statistics in two sections Overall Model and Individual Factors The components of this worksheet are presented below Analysis Statistics t Tests Overall Model Individual Factors Statistics Value Statistics Minority Int a 1 R 0 09 b 3815 73 18414 38 Sig p 0 00 t value 5 87 57 85 Valid n 363 00 Sig p 0 00 0 00 2 R 0 18 b 1964 41 28211 36 Sig p 0 03 t value 2 34 48 44 Validn 27 00 Sig p 0 03 0 00 3 R 0 56 b 24965 12 49314 03 Sig p 0 00 t value 10 48 72 99 Valid n 87 00 Sig p 0 00 0 00 1 Overall Model Interpretation of the Overall Model statistics provides insight into the strength of the specified model e R Sar R Provides an estimate of effect size i e the amount of variance accounted for by the model e Sig p Provides the p value for the statistical
24. erformance Appraisal Score s 4 Comparison Group Variable Required This is one of two required specifications and is critical to specifying the correct compensation analysis This variable identifies members of groups 0 Focal 1 Reference to compare in the compensation model From the Available Variables box select the grouping variable and click the right arrow button gt next to the Comparison Group Variable box Note that only the Binary lt BIN gt datatype is permissible 5 Analyze By Variable s Optional Specifying the Analyze By variable is optional but it should not be ignored if employees in the data are grouped into Similarly Situated Employee Groupings SSEGs COMPARE can easily run multiple analyses simultaneously e g by department job title etc Currently the program can handle up to five strata However based on previous experience there is rarely ever a need to overlap more than three strata For example a typical three strata specification would be AAP Location x Job Group x JobTitle To specify the analysis strata simply select each SSEG layer variable and click the right arrow button gt next to the Analyze By Variable s box It is important to note that the order in which variables are placed into this box will determine the way in which the results are displayed For example if Location is selected first followed by Job Group the results will display all Job Groups within each Loc
25. f this worksheet are presented below Descriptives Report Group Da Focal Reference Gand ped DI parenie 1 Count of Minority 276 87 363 Focal 17 16 High Avg of Salary 18 414 38 22 230 11 19 328 90 3 815 73 Avg of PPA 2 0 1 8 1 9 Avg of TIC 4 7 17 1 Tad Avg of TIJ 4 6 16 9 TS 2 Count of Minority 14 13 27 Focal 6 51 Medium Avg of Salary 28 211 36 30 175 77 29 157 19 1 964 41 Avg of PPA 3 2 3 0 3 1 Avg of TIC 11 1 28 2 19 4 Avg of TIJ 2 9 29 2 15 5 3 Count of Minority 80 7 87 Focal 33 61 Medium Avg of Salary 49 314 03 74 279 14 51 322 71 24 965 12 Avg of PPA 4 2 ST 4 4 Avg of TIC 6 8 28 1 8 6 Avg of TIJ 5 2 16 3 6 1 Data Indicates the type of descriptive statistics computed o Indicates which group is negatively impacted Analyze By SSEG Represents the analysis groups selected by the analyst e g Job Group Focal Reference Provide descriptive statistics for Focal and Reference group members Grand Total Provides summary descriptive statistics for the entire group Focal Reference Impacted Difference Provides two pieces of high level information o Provides the raw mean difference between Focal and Reference group members Difference Provides the percent difference in compensation between Focal and Reference group members The higher paid group is used as the computed reference 19 20 Potential Liability Indicates whether the potential for liability is Low Medium or High based
26. gression 17 18 o Signif t Test but Non Signif Regression Provides a count and percentage of the number of SSEGs that were significant based upon t Test results but became non significant when using multiple regression o Workforce in Groups with Signif t Test but Non Signif Regression Provides a count and percentage of the workforce in SSEGs where pay disparities were significant when based upon t Test results but became non significant when using multiple regression Liabilities This category provides several high level summary of liability by statistical model t Test regression and group membership Focal Reference o Total Liabilities t Test Focal Provides the total dollar amount for Focal group members e g women minorities etc necessary to eliminate the statistically significant differences in compensation summed across all analyses based on Test results o Total Liabilities Regression Focal Provides the total dollar amount for Focal group members e g women minorities etc necessary to eliminate the statistically significant differences in compensation summed across all analyses based on multiple regression results o Total Liabilities t Test Reference Provides the total dollar amount for Reference group members e g women minorities etc necessary to eliminate the statistically significant differences in compensation summed across all analyses based on t Test results o Total Liab
27. hese detailed reports can select them from the Output Reports options 2 Qutlier Analysis COMPARE identifies outliers by computing their standardized z values to determine how far a value may be from the mean By default COMPARE only flags outliers that exceed 1 96 standard deviation units without specifying the exact z value Analysts who wish to evaluate the actual z scores may obtain them by selecting the Outlier Analysis option 3 Group Heading Specifications By default COMPARE labels the group headings as either Focal or Reference Analysts may choose to customize their reports by entering more specific labels based upon the analyses being conducted 4 Data Worksheets In the output reports workbook COMPARE does not automatically include the analysis dataset or copies ofthe original data Ifthe analyst desires to replicate the analysis data in the output workbook it is possible to request copies of the original data e Analysis Data is the subset of variables and values specified in Step 5 Analysis Setup e Copy of Original is a copy of the original dataset in its entirety 5 Report Header COMPARE includes an Attorney Client Privileged header across all reports The analyst may change this header by editing the field though it is recommended to leave it in place 11 Step 7 Configuring the COMPARE Analysis Specifications On the Analysis Specifications screen the user can cus
28. iables specified in the model The components of this worksheet are presented below Analysis Statistics Regression with Explanatory Variables Overall Model Individual Factors Statistics Value Statistics Minority TIJ TIC PPA Int a iR 0 81 b 5493 63 56 57 3 26 5551 13 7667 90 Sig p 0 00 t value 9 07 0 77 0 05 36 71 19 82 Validn 363 00 Sig p 0 00 0 44 0 96 0 00 0 00 R 0 65 b 10139 66 471 75 4 57 1042 27 23564 23 Sig p 0 00 t value 3 20 3 48 0 04 4 38 15 81 Valid n 27 00 Sig p 0 00 0 00 0 97 0 00 0 00 3 R 0 91 b 6234 18 122 30 1236 39 2359 70 30242 01 Sig p 0 00 t value 1 26 0 94 5 26 8 32 22 29 Valid n 87 00 Sig p 0 21 0 35 0 00 0 00 0 00 1 Overall Model Interpretation of the Overall Model statistics provides insight into the strength of the specified model e R Sqr R Provides an estimate of effect size i e the amount of variance accounted for by the model e Sig p Provides the p value for the statistical test of significance 1 e whether the model is a significant predictor of compensation p values less than or equal to 0 05 are generally considered statistically significant e Valid n Provides the sample size of the analysis 2 Individual Factors Interpretation of the Individual Factors provides traditional regression coefficient statistics e b b Provides the unstandardized regression coefficients they can be used to determine the directionality and the strength of the
29. ilities Regression Reference Provides the total dollar amount for Reference group members e g women minorities etc necessary to eliminate the statistically significant differences in compensation summed across all analyses based on multiple regression results o Liability Excluded Provides the number of focal group members that are excluded from the regression and thus the liabilities calculated by COMPARE For example if 50 women are included in the analysis and two explanatory variables are included in the model e g tenure and job performance scores but complete data is only available for 40 of the women the program will display the error noted in the Results Summary above When this error appears Full Information Maximum Likelihood FIML statistical modeling should be used to impute data for the missing records Call BCG for assistance Important Note It is strongly recommended that no adjustments in pay be made without first validating the model and exploring whether other factors that were not included in the model but which are otherwise legitimate may be influencing pay It is recommended that analysts and employers seek qualified statistical assistance at this stage Worksheet 2 Descriptives The Descriptives worksheet presents descriptive statistics for each of the variables included in the analysis broken down by each level of the Analyze by Variable SSEGs selected during Analysis Setup The components o
30. luded liabilities is X a problem has occurred with the analysis where some data for one or more explanatory variables for the at issue group is missing When this error appears 32 Full Information Maximum Likelihood FIML statistical modeling should be used to impute data for the missing records Call BCG for assistance Important Note COMPARE is a powerful and comprehensive statistical tool that takes advantage of the advanced statistical technique known as multiple regression analysis to explore company pay practices It is particularly useful in determining whether differences in pay are due to potential discrimination or due to legitimate job related factors There is always a danger when automating such a powerful technique that steps will not be taken to ensure that the analyses comply with good scientific practice Therefore Biddle Consulting Group Inc strongly encourages users who are not highly experienced with advanced statistics to seek some level of oversight to ensure that the results of analyses are valid reliable and defensible Those who use COMPARE do so at their own risk and Biddle Consulting Group Inc makes no warrantees express or implied with respect to the findings or liabilities resulting from its use When data has certain ordered properties it may be coded into numeric format e g Best 2 Acceptable 1 Worst 0 This is an advanced procedure and requires an analyst with sufficient experien
31. lues on a variable These employees can have a dramatic impact on the results of an analysis and can lead to misleading findings of compensation disparities or the lack thereof Skewness Skewness is caused when outliers tend to exist in only one direction either the outlier s are all high or are all low COMPARE computes a skew index and determines whether it is significant or not Only significantly skewed variables are flagged o Skew Indicates significant positive skewness If positive skewness exists it means there is a group of employees with higher values on a variable than the typical group of employees Skewed data can lead to misleading results o Skew Indicates significant negative skewness If negative skewness exists it means there is a group of employees with lower values on a variable than the typical group of employees Skewed data can lead to misleading results o Small N Indicates that the number of employees in the group is too small for a valid calculation of skewness Kurtosis Kurtosis is a measure of variation which indicates the extent to which employees are spread out on a variable COMPARE computes a Kurtosis index and determines whether it is significant or not Kurtosis is flagged only if it is significant 23 24 Kurt Indicates significant positive Kurtosis which is a tight peaked distribution Kurtosis in the data may mean a violation of normality required by multiple regression analyse
32. lyses become difficult and the potential liability is decreased The default labeling is Low 3 Medium 6 High 10 e Group Diff This is the difference in average salary between Focal and Reference groups The larger the difference the greater the potential liability The default is to flag groups with differences greater than 5 Multicollinearity COMPARE provides two measures of multicollinearity VIF and Pearson s r to identify correlations between variables The default thresholds reflect well established statistical standards Outlier COMPARE identifies outliers by computing the z score for each value by variable The default threshold of z gt 1 96 will flag values that are statistically significantly different than the mean i e they exceeds the 95 confidence interval around the mean 3 Sample Size Requirements For statistical tests to provide valid conclusions the sample sizes need to be sufficient COMPARE evaluates sample size on two levels e All Total Total number of records 1 e employees in each SSEG e Subgroup The number of employees in each subgroup e g men women within an SSEG SSEGs that do not meet these very low sample size requirements for an analysis will not be analyzed After specifying the Report Setup and Analysis Specifications return to the Analysis Setup screen to run the analysis by clicking on the L button Important Note Changes to the default settings withi
33. n COMPARE are not saved If the default settings are changed they will need to be changed again for each subsequent set of analyses Step 8 Running the Analysis Once the compensation model has been specified i e variables moved to the appropriate boxes and the reports have been selected and configured the next step is for COMPARE to run the battery of analyses and generate statistics that will help evaluate whether the model is statistically significant whether there are unexplained differences in pay by group e g gender after controlling for legitimate job related explanatory factors and where pay adjustments may be necessary 13 To run the analysis simply click Run on the Analysis Setup screen If the data set is large or ifthere are many different strata i e multiple split levels the analysis may take some time to complete The Analysis Status window should indicate the progress of the analysis If a problem occurs please notify COMPARE technical support support biddle com Analysis Summary and User Warning Once COMPARE completes all computations and reports the results are summarized in a high level overview Compare 2 x Biddle Consulting Group Inc Copyright 2009 xj The Analysis Summary and User Warning screen has four Model Evaluation CATEGORY PROBLEM SEVERITY DESCRIPTION major CO lumns Descriptives Yes 30f3 Disparity Medium and High Correlations No Signif Overall
34. nt from the typical employee This information will be useful when evaluating whether all employees in the SSEG are similarly situated outliers will be flagged in red or in the event a cohort analysis becomes necessary as a result of statistically significant multiple regression findings 25 Worksheet 7 Interaction A statistical interaction occurs when an explanatory variable works differently for one group versus another These relationships are investigated in the Interactions worksheet The components of this worksheet are presented below Interaction Report Grade Job Code Interaction R2 Change r partial t Value p Value 1 Job ABC Minority x PPA 0 014 0 226 4 533 0 000 Minority x TIC 0 001 0 033 0 654 0 514 Minority x TIJ 0 003 0 059 1 159 0 247 The primary role for COMPARE is to simplify the process of conducting a valid multiple regression analysis for those who may not be professional statisticians but who are required to use this statistical technique to evaluate compensation Specifically COMPARE tests for interactions between the Test Group Variable i e Focal Reference gender minority status and each specified Explanatory Variable by SSEG The user can either run one overall interaction report for the entire dataset or individual interactions for each SSEG COMPARE summarizes the results in the Interaction worksheet There are at least five columns on the Interaction report 1 2 Optional Gro
35. o the next tab Each of these tabs will be discussed below WARNING Carefully review and understand each tab in the output workbook from left to right before moving to the next Review all tabs before attempting to interpret the analysis results Interpretation of and descriptions for each of the report worksheets are presented in order in the following sections 15 Section 1 Evaluating the Model Worksheet 1 Summary The Summary worksheet is the core of the program because it quickly presents the analyst with warnings and a summary regarding whether problem areas were found with employee compensation There are two primary sections to this report Model Evaluation and Results Summary The components of this worksheet are presented below Model Evaluations Category Problem Severity Description Descriptives Yes 3 of 3 Disparity Medium and High Correlations No Signif Overall Pay Disparity Multicollinearity Yes High r V1 V2 gt 0 8 No VIF gt 10 Diagnostics No Missing Data Yes 107 Outliers Outliers Yes 1 Skew Skewness Yes 1 Kurtosis Kurtosis 3 3 100 Groups Analyzed with t Test 477 477 100 Workforce Analyzed with t Test 3 3 100 Groups Analyzed with Regression 477 477 100 Workforce Analyzed with Regression Interaction Yes 1 3 33 Potentially Tainted Variable e Descriptives YES indicates COMPARE identified one or more areas of medium or high potential liability
36. of compensation absent any explanatory variables p values less than or equal to 0 05 are generally considered statistically significant Reg p value Provides the p value for the multiple regression analysis i e whether the protected variable is a significant predictor of compensation after controlling for any explanatory variables p values less than or equal to 0 05 are generally considered statistically significant If NC is indicated in the results it means that the statistics could not be calculated either due to small sample size or because there were no members of one of the comparison groups This is a useful report that may be printed and used during discussions with management 31 Worksheet 11 Liabilities In the typical analysis Liabilities is the last worksheet This report provides an estimate of the additional pay that would need to be allocated to specific employees to reduce observed pay disparities to the desired standard deviation SD difference default is 1 SD Computed liabilities are presented in the last two columns of this report The components of this worksheet are presented below WARNING Do not make adjustments to employee compensation without first determining whether other factors could be influencing pay in a legitimate manner and that the multiple regression model itself is valid and reliable Liabilities Report Job Emp ee a es ID Salary PPA TIC TIJ Minority t Liable R Lia
37. orkbook click the down arrow next to the Output Workbook box and click New Then click Select The results will be placed into a new Excel workbook 2 Saving Results to an Existing and Opened Workbook To save the results into an existing workbook that workbook must be opened at this point for it to be listed in the drop down selection box Click the down arrow next to the Output Workbook box and select the name of the workbook into which to place the analysis results Then click Select Once the Data Input Source and Output Workbooks have been selected click the Next button Step 5 Configuring the COMPARE Analysis Setup This screen is where the analyst specifies the compensation model i e the variables that belong in the compensation analysis 7 p y Compare 2 x Biddle Consulting Group Inc Copyright 2009 Analysis Setup There are six major fields on Analysis Setup hi Available Variables Salary this screen lt NUM gt ID lt STR gt Birthdate E gt 1 Available Variables lt NUM gt Educ Explanatory Variables lt STR gt Bunit lt NUM gt JobGroup 2 Salary lt NUM gt Salary gt lt NUM gt SalaryStart 3 Explanatory lt NUM gt PPA 9 lt NUM gt PPA22 Comparison Group Variable Variables er 3 lt NUM gt TIJ 2 4 Comparison Group BI RAR SER E Analyze By Variable s Variable lt BIN gt Gender gt 5 Analyze By gt Variable s Employee I
38. relationship between the specific variable and compensation e f value Provides the Test statistics e Sig p Provides the p value for the statistical test of significance i e whether the model is a significant predictor of compensation p values less than or equal to 0 05 are generally considered statistically significant 29 Worksheet 10 Compensation Analysis The Comp Analysis worksheet provides a powerful summary of the compensation analysis in a very compact and efficient format At a glance the analyst can identify which SSEGs were significant and the severity of the pay disparity The components of this worksheet are presented below Compensation Analysis Results _ Reg Impact t Test Reg Hh Data Focal Reference an i ji a f Group Tota Diff Diff p Value p Value 1 Count of 276 87 363 Minority Average 18 414 38 22 230 11 19 328 90 of Salary Average 2 0 1 8 1 9 of PPA Average 4 7 17 1 7 7 of TIC Average 4 6 16 9 7 5 of TI 2 Count of 14 13 27 Minority Average 28 211 36 30 175 77 29 157 19 of Salary Average 3 2 3 0 3 1 of PPA Average 11 1 28 2 19 4 of TIC Average 2 9 29 2 15 5 of TI 3 Count of 80 7 87 Reference 0 21 Minority Average 49 314 03 74 279 14 51 322 71 6 234 18 of Salary Average 4 2 5 7 44 of PPA Average 6 8 28 1 8 5 of TIC Average 5 2 16 3 6 1 of TIJ The sample report above indicates 30 JobGroup 1 The t Test indicates the pay disparity n
39. reports The results from this analysis are based on advanced amp complex statistical models BCG reflect the principle that the advises the user to consult an expert www Biddle com for appropriate interpretation prior sa to making any compensation decisions The user assumes all responsibility for use and validity of the compensation interpretation of these results Ina Sao model should be evaluated first Proceus Timer 4LE7 g g and the results second The Analysis Summary and User Warning screen has two major sections Model Evaluation and Results Overview 1 Model Evaluation Lists any problems with the data identified during diagnostics that may undermine the validity of the statistical analyses The model evaluation summary includes warnings for Potential Disparity Significant Correlations Multicollinearity Data Diagnostics e g outliers skewness kurtosis Proportion of Data Analyzed and Interactions 2 Results Overview Lists a summary of significant and non significant t Tests and regression analyses and total liabilities by statistical model Important Note It is critical that this screen be carefully reviewed COMPARE was designed with an efficient and user friendly interface with the intent to allow non statistical analysts to run very complex statistical analyses However unless the data is accurate assumptions are met and the results have been validated no conclusions can be made with confidence
40. s Kurt Indicates significant negative Kurtosis which is a wide low distribution Kurtosis in the data may mean a violation of normality required by multiple regression analyses Small N Indicates that the number of employees in the group is too small to allow for a valid calculation of kurtosis Worksheet 6 Outliers The Outliers worksheet provides a detailed look at specific employees identified in the Diagnostics worksheet Those identified as having values on a variable that are either unusually high or low compared to other employees are highlighted The components of this worksheet are presented below Outliers Report JobGroup Emp ID Salary PPA TIC TIJ Minority 1 376 17 416 2 0 8 0 5 0 0 28 17 179 2 0 7 0 5 0 0 43 15 832 1 0 1 0 3 0 0 36 17 546 2 0 10 0 8 0 0 369 16 713 2 0 4 0 2 0 0 56 17 055 2 0 2 0 3 0 0 90 14 286 1 0 3 0 5 0 0 91 15 809 1 0 0 0 3 0 0 472 25 821 3 0 7 0 6 0 0 5 29 380 7 0 8 0 0 17 2 0 1 0 0 52 29 591 4 0 1 0 0 80 0 0 3 0 0 234 29 435 3 0 1 0 0 310 30 122 5 0 8 0 0 319 28 010 1 0 1 0 0 12 3 0 4 0 0 Without clear justification it is recommended that the analyst not attempt to remove outliers from the data because it could be seen as manipulating the data in a way that gives the results a company desires Rather each outlier should be carefully examined to determine whether there are legitimate reasons for their scores to be so differe
41. scription of this method 12 2 Warning Threshold Specifications COMPARE has built in features to highlight and flag areas of potential concern The default settings for triggering these warnings are based on best practices and statistical standards Advanced analysts may change these thresholds to better suit their needs For most cases however it is advisable not to change the defaults especially if the analyst has no clear justification to change the sensitivity of these tests Potential Liability COMPARE provides a very powerful and sensitive test to help identify potential liabilities It is descriptive based rather than statistical but is very helpful in identifying areas that may trigger further investigations from federal enforcement agencies Groups flagged with a medium or high potential liability should be given the most attention COMPARE flags three levels of potential disparity Low Medium and High by evaluating three descriptive factors e All Total This is the total number of individuals in each SSEG The larger the number of employees the higher the likelihood of potential liability if there are differences in average compensation The default labeling is Low 10 Medium 20 High 30 e Subgroup The number of individuals in each subgroup e g men women within each SSEG Even if there is a large number of total employees within the SSEG if there are not many in both groups then statistical ana
42. sis cccccccessceecesseeeseeeseeeeeceecsaecaecnseceaeeeeeeseeeseneeaes 13 IV Interpreting Compensation Analysis Results in COMPARE 15 Section 1 Evaluating the Model unnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn 16 Worksheet 1 Summary oo cccccccccseesseceteceseceseceseceeeeseeeceeeeeseeeseecseecseeeaecsaeenseeeaeees 16 Worksheet 2 Deschiptives u a ninia iaa 19 Worksheet 3 Correlations uusesseenseenseenseennnennnennnensnnnsnnnnnennnnnsennse esse enn ernennen 21 Worksheet 4 Multicollinearity ccccccccsccssccsseceseceseceeeceeeeseeeeeseeeeeeeseeeseeeseeesaees 22 Worksheet 3 Diagnostics iii id didas bdo 23 Worksheet 6 Outliers 0220sussensennenssensennsnnnnsnensonnnsnennennnsnennennennnsnenenn 25 Worksheet 7 Interaction uueesssesseessensennennnennnennnnnnnnnenennennennsennseenseensens nn 26 Section 2 Evaluating the Statistical Results oooonnnnnccnnonennnno 28 Worksheet 8 Stats No Exp Var cccccccccccsseesteceteceseceeceeceeeeseeeeseecaeeeeenaeenseeeaeens 28 Worksheet 9 Stats Exp Var eeseensenseensnessneesnennnnesnnnnnnnnnennennsennse esse iniia biaen 29 Worksheet 10 Compensation Analysis ccccescceseessceeseeeeeeseeeeeeeeessecnseenseeeaeens 30 Worksheet 11 Liabilities oooononnnnnononanoconnnnonnananenocnonononnanononocnononannnnonecicnononanns 32 I Conducting and Interpreting Analyses in COM
43. sly integrate with Microsoft Excel The program was developed from the ground up as a suite of stand alone modules assembled together in an Excel workbook Installation is simply a matter of opening the COMPARE Excel workbook the same way any other Excel file would be opened Activation is required after the user has downloaded COMPARE from www BCGinstitute org or installed it from the CD The program will provide an Activation ID that must be registered for COMPARE to operate Minimum System Requirements Minimum installation requirements include e Microsoft Windows XP Vista or 7 e Microsoft Excel 2003 or 2007 Important Note Depending on your security settings you may receive a warning that COMPARE requires that Excel macros be enabled Some organizations set their security settings so that macros are disabled by default to protect against viruses or malicious software The user can adjust Excel to permit COMPARE to run but not other macros If you have additional questions contact your organization s Information Technology IT Department or Technical Support at Biddle Consulting Group at 1 800 999 0438 111 Conducting Compensation Analyses in COMPARE There are eight steps to running a complete compensation analysis in COMPARE The following is a brief overview Details will be presented in subsequent sections 1 Prepare the Data Prepare the data to the specifications detailed in the Data Preparation section
44. test of significance i e whether the variable is a significant predictor of compensation p values less than or equal to 0 05 are generally considered statistically significant e Valid n Provides the sample size of the analysis 2 Individual Factors Interpretation of the Individual Factors provides traditional t Test statistics e b b Provides an estimate of mean group differences e t value Provides the Test statistics e Sig p Provides the p value for the statistical test of significance 1 e whether the model is a significant predictor of compensation p values less than or equal to 0 05 are generally considered statistically significant Because no explanatory variables are included in this model the analyst should interpret the at issue variable which in this example is Individual Factors Minority The intercept nt a is largely ignored but may provide valuable insight for the more advanced analyst 28 Worksheet 9 Stats Exp Var The raw regression statistics are presented in the Stats Exp Var Explanatory Variables worksheet This is the full regression model with all explanatory variables Proper interpretation of the report requires a firm understanding of regression statistics The report presents the statistics in two sections Overall Model and Individual Factors The columns in the worksheet are ordered such that the at issue Test Group Variable is presented first followed by all explanatory var
45. the Input Data Source and the destination for the output reports Input Data Source Compare 2 x Biddle Consulting Group Inc Copyright 2009 Users must specify the Excel Data Setup Workbook and Worksheet of the data to be analyzed 1 Select Workbook Select Workbook Compare 2x xism Select the workbook containing the data to be analyzed Note that the workbook must already be open as instructed in Output Workbook Step 2 To select the workbook simply click the drop down arrow next to the Select Workbook box Next click the Select button to instruct COMPARE that the workbook has been selected 2 Select Worksheet Select the worksheet containing the data to be analyzed Note that the only worksheets available to select are those contained in the workbook previously selected If there are multiple sheets in the workbook be sure to select the worksheet containing the data that has been specifically prepared for analysis To select the worksheet click the drop down arrow next to the Select Worksheet box Next click the Select button to instruct COMPARE that the worksheet has been selected Output Workbook Input Data Source Select Worksheet Demo Select Workbook New The final step of data setup is to select a workbook into which the results of the analysis will be written Two options exist 1 Saving Results to a New Workbook To save the results to a new w
46. tomize detailed analysis settings There are three major specification groups 1 Liabilities Analysis 2 Warning Thresholds and 3 Sample Size Requirements Analysis Specifications f Compare 2 x Biddle Consulting Group Inc Copyright 2009 3 1 Liabilities Analysis The Report Setup Analysis Specifications following are the default Liabilities Analysis specifications Liabilities Distribution 5 Evenly Distribute All Individuals e Compute Liabilities COMPARE computes liabilities by default If the Warning Threshold Specifications analyst does not Potential Liability wish to generate Low Medium High cetera it gt All Total 2 Liabilities select All Total 10 20 30 otal 10 o No Subgroup 3 6 10 Subgroup 3 s Evenly Distribute Below Average Proportionally Weighted by Impact Sample Size Requirements t Test Regression e Standard Deviation Be E gt El Diff By default Multicolinearity Outlier liabilities are VIF gt r gt z gt computed to reduce 10 0 90 1 96 the pay disparity to one standard deviation difference For the advanced analyst it is possible to modify this by selecting from the drop down menu or entering a specific value e Compute Reference Group Liabilities Compensation analyses evaluate impact against both Focal and Reference group members This is particularly relevant as Title
47. up by Identifies the grouping variables e g job group job title etc by which the compensation analyses will be run 3 Interaction Identifies the interaction that is evaluated 4 R2 Change Identifies the change in the multiple correlation AR which is the incremental variance accounted for by the interaction term This is valuable information for the advanced analyst who seeks to evaluate the severity of the interaction 5 r partial Provides the partial R of the interaction term as unique incremental variance This is valuable information for the advanced analyst who seeks to evaluate the severity of the interaction 6 t value Provides t value for the statistical test of significance 7 p value Provides the p value for the obtained t value Values in this column less than or equal to 0 05 indicate a significant interaction Significant interactions are highlighted in red Once a significant interaction is identified the analyst should investigate the variable that is significantly interacting with the Test Group Variable e g gender race etc and compensation This is a complex topic A thorough discussion of the topic is beyond the scope of this manual With that said interpretation often requires a graph of the interaction See below for an example of an interaction 26 Interaction Between Annual Compensation and Performance Appraisal Score 65 000 O 60 000 re A O E 55 000 3
Download Pdf Manuals
Related Search