Pleading

I, Dr. Richard Drogin declare:

I make this declaration of my own personal knowledge and could testify thereto if called as a witness.

A. INTRODUCTION

1. I have been retained by Plaintiffs’ counsel to analyze statistical questions raised in the Dukes v. Wal-Mart Stores, Inc. litigation. The purpose of my study was to obtain descriptive summaries of computer data, and prepare various statistical analyses relevant to the issues in the case. I have previously submitted a report in this litigation, dated February 3, 2003. Subsequently, defendant’s expert Dr. Joan Haworth submitted a report dated April 1, 2003 giving her opinion and results regarding her review of my February 3 report. On April 18, 2003 she provided an amended report and revised analyses. Then, on April 21, 2003, at her deposition, she provided additional backup materials for her revised analyses. I attended that deposition. Subsequently, she filed a declaration in Opposition to Class Certification. The declaration below gives my rebuttal to Dr. Haworth’s report and deposition testimony. Except for the addition of citations to Dr. Haworth’s declaration in Opposition to Class Certification, the attachment of cited documents, and the correction of typos, this declaration is identical to the rebuttal report I wrote dated May 6, 2003.

2. Dr. Haworth’s report covers several issues regarding plaintiff’s expert reports. My rebuttal will address those areas where she has presented her opinions regarding issues covered in my February 3 report. In this declaration I will explain the flaws in her conclusions regarding the promotion process and compensation system at Wal-Mart. Unless otherwise noted, all references given below to her report will refer to the original report dated April 1, 2003 and corresponding references in her Declaration. She has testified at her deposition that there are no substantive differences between her April 1, 2003 report and the revised report of April 18, 2003. Her declaration adds additional analyses and material, which I have not analyzed. Some of the tables in her Declaration also cover a shorter time frame than those utilized in her report, and thus the numbers do not always correspond to those in her report. My rebuttal report, and this declaration, do not address these changed numbers.

B. SUMMARY OF FINDINGS

3. Dr. Haworth’s analysis and conclusions are defective because she:

a. Relies on incomplete and "nearly useless" job posting data in her analysis of promotions to hourly jobs;

b. Fails to note that job posting selections contribute to gender segregation by department;

c. Fails to note that the average percentage of women in the appropriate pool of most qualified applicants determined from the MIT program is nearly identical to the average percentage of women in the availability pool I used in my promotion analysis presented in my February 2003 report, thus corroborating my earlier analysis of promotions into MIT;

d. Relies on Management Career Selection data that is incomplete and biased, thereby ignoring all promotions to salaried jobs except Store Manager;

e. Inappropriately disaggregates hourly employees into 7500 separate subunits to perform her regression analysis of hourly pay rates because she:

· Ignores documented company policy for setting pay rates in her decision to analyze stores separately;

· Incorrectly applies statistical methodology in deciding to analyze stores separately;

· Arbitrarily subdivided stores further into smaller grocery/non-grocery and specialty subunits, which is unsupported by any written company policy;

f. Relies on a Store Manager Survey to justify her regression model, despite admitting that the survey methodology violates accepted scientific standards, and was conducted by defense counsel contrary to her recommendations;

g. Erroneously excludes from her hourly pay rate analyses persons who were Department Heads;

h. Arbitrarily and inexplicably excludes from her hourly pay rate regressions persons who were ever demoted and persons who ever had been salaried;

i. Includes tainted variables in her compensation regression analyses despite her testimony in other litigations stating that this should not be done;

j. Includes variables in her compensation regressions without justification or explanation;

k. Fails to report statistically significant aggregated results of her subunit regressions even though she has testified in other litigations that such aggregation is appropriate to determine if there is an overall disparity;

l. Fails to report compensation regressions for salaried employees other than Store Manager even though these are included in her backup materials.

C. PROMOTIONS

4. Dr. Haworth criticized my promotion analyses, and presented various analyses of her own derived from the Job Posting System, Management Training Posting, and Management Career Selection System. Each of these is addressed below.

Job Posting Data – Nearly Useless

5. Dr. Haworth presents a lengthy discussion of the job posting system at Wal-Mart and does several analyses concluding that “[w]hen all of the job postings for all the stores in all districts and all regions are all aggregated there are 3,266 more female job offers than expected in a gender-neutral process when controlling for department and job code.”^{^[1]} Dr. Haworth’s analysis and conclusion are misleading, and not probative, because the job posting data upon which she relies is incomplete and is not utilized systematically.

6. For example, for promotions into Support Manager, a job where it is possible to determine the completeness of the job posting data relative to actual promotions recorded in the PeopleSoft job history data, the job posting data includes posting and acceptances in only 20% of the total number of actual promotions found in the PeopleSoft data.^{^[2]} With such a large percent of vacancies filled outside the job posting system, and no policy or explanation regarding when the system was utilized or by-passed, no meaningful analysis of promotions can be conducted from this dataset.

7. Dr. Haworth relied on the job posting data, but made no attempt to evaluate or study the completeness of the job posting data.^{^[3]} At her deposition, when asked whether she had done any “... analysis for any position in the job posting data versus Global PeopleSoft”, she answered, “I don’t recall doing such.”^{^[4]} Further, when asked “So you don't know what proportion of positions that were filled in the Global PeopleSoft were filled by posting for any hourly job?” she answered “...I do not know what proportion of the moves into support manager, whether demotions, promotions or laterals, were therefore covered by the posting.”^{^[5]}

8. In fact, it can only be concluded that the job posting data is “nearly useless” according to Dr. Haworth’s published statements in the Employee Relations Law Journal. Dr. Haworth’s failure to note the high degree of incompleteness of the job posting data, the apparent lack of any system governing when the job posting was used, and her failure to present any study of these obvious problems directly contradict the first caveat she espoused in an article co-authored with her husband published in the Employee Relations Law Journal:^{^[6]}

“The information collected from applicants must be sufficient to allow a proper analysis of the selection process in a race- and gender-neutral environment. To this end, there are two general caveats. First, information that is collected but never verified or checked for accuracy is nearly useless.”

Job Posting Selections Contribute to Gender Segregation by Department

9. Ignoring the defects with job posting data described above, Dr. Haworth fails to note, or present any explanation of, obvious gender segregation patterns indicated by her job posting promotion analysis. For example, her analysis shows that female bidders for Department Head jobs receive significantly more promotions in the departments with the highest percent female than would be expected based on their application rate, and significantly fewer promotions in departments with highest percent men. Tables 1a and 1b below list the ten departments with the highest percent women, and the lowest percent women^{^[7]} as of year-end 2001.

Table 1a

Dr. Haworth Job Posting Analysis for

10 Departments with Highest Percent Women

Target Department % Wom Diff. Z-Value

34 Ladies Sportswear 99.2 35.10 5.20

27 Hosiery 99.1 66.30 6.90

19 Piece Goods 99.1 61.50 7.40

46 Health & Beauty 98.7 56.40 6.70

26 Infants & Toddlers 98.6 79.90 6.60

32 Jewelry 97.3 58.30 6.20

910 Back Office 94.2 203.20 9.30

23 Men's Wear 92.5 92.80 6.90

20 Domestic Goods 92.4 20.90 1.90

40 Pharmacy 88.8 54.20 5.00

Total 728.60 19.41

Table 1b

Dr. Haworth Job Posting Analysis for

10 Departments with Lowest Percent Women

Target Department % Wom Diff. Z-Value

16 Horticulture 39.3 8.30 0.40

8 Pets and Supplies 37.6 -77.00 -3.10

284 Div 28 Receiving 30.7 7.50 1.40

9 Sporting Goods 30.2 14.20 0.70

4 Paper Goods 29.5 -33.40 -1.30

11 Hardware 27.7 -26.90 -1.20

90 Dairy Products 25.9 -30.20 -2.70

93 Meat 21.9 -4.10 -0.70

94 Produce 9.4 -10.70 -1.90

37 TBO Service 6.7 -170.70 -3.60

Total 170.00 -3.18

10. Tables 1a and 1b show that women received 728.6 more offers for Department Head jobs in highly female departments than expected, and 170 fewer offers than expected in highly male departments, based on their percent among applicants. These disparities are statistically significant: For the over-promotion of women into Department Head in highly female departments with Z-value of 19.41, there is only 1 chance in 10 to the 70th power that a disparity this large would occur under random selection. For the under-promotion of women into Department Head in departments with the lowest percent female the Z-value of –3.18 indicates there is less than 1 chance in 700 that a disparity this large would occur under random selection.^{^[8]} For both of these analyses, the expected number is based on the percent of women among applicants who applied for the positions, as determined from the job posting data. Thus, Dr. Haworth’s job posting analysis demonstrates that gender segregation by department is perpetuated, in part, through the job posting system.

2003 Assistant Manager Training Program Corroborates Plaintiff’s Analysis

11. Prior to January 2003 Wal-Mart had no system for hourly employees to express interest or apply for any entry level salary management positions. In January 2003 Wal-Mart introduced for the first time a system for accepting applications for their new Assistant Manager Trainee Program (also referred to as the Management-In-Training Program, or, simply, MIT Program). During a one week period Wal-Mart received about 30,000 applications through this system, and about 1400 selections were made during March.

12. Dr. Haworth presents an analysis of the data from this MIT Program, and reports that “The percentage of females who voluntarily expressed interest in 2003 in promotion to management levels (44%) is similar to the 41% women were among those who were promoted in the Assistant Manager Trainee positions in the five years prior to the inception of this program.”^{^[9]} Her statement (“44% is similar to 41%”) suggests that the 2003 MIT program indicates the selection of female hourly employees to positions as MIT during 1998-2002 was consistent with their interest and qualifications for such positions.^{^[10]} In fact, the MIT data she presents shows the exact opposite, and corroborates the analysis of promotions into MIT positions that I presented in my report.

13. Assume, as Dr. Haworth does, that the January 2003 MIT program was fair and unbiased with respect to gender^{^[11]}. Under this assumption, the data from the MIT program might be used to obtain the correct unbiased percentage of women among the most qualified hourly employees who want to enter salary management positions. Dr. Haworth incorrectly suggests the fair availability percentage would be 44%, i.e. the percentage of women among all those applying in the MIT program. However, the correct percentage of females among the “true, unbiased” availability pool would be those interested and available, and also most qualified. There may be many men and women who submitted applications in the 2003 MIT program who expressed interest in promotion, but for some reason are not among the most qualified. The group of people who are interested, available, and most qualified would be determined by the percentage of females among those actually selected from the process, since presumably (as Dr. Haworth believes^{^[12]}) Wal-Mart has selected the best, most qualified applicants. The percentage of women selected in the MIT program was 59.8%. Thus, based on Dr. Haworth’s assumptions, the most accurate availability figure of women who are interested in MIT positions among those most qualified would be 59.8%.

14. In my February 2003 report I presented an analysis of promotions into MIT positions during 1997-2002. I determined the percentage women in the pool of those available for promotion to be 59.6%, as shown in Table 23 of my report. My determination of this percentage was based on the percent of women in the historical feeder pools for the MIT positions.^{^[13]} The 59.6% availability figure I derived in my promotion analysis is nearly identical to the 59.8% availability figure derived from the January 2003 MIT Program selections, as described in the previous paragraph. Based upon my analysis, I found that there was a shortfall of about 3000 females promoted to MIT positions during 1997-2002. Since the 59.8% and 59.6% availability figures are so close, the results of the recent 2003 MIT Program corroborates the female availability for promotions into MIT positions I used to compute this shortfall in female promotions to MIT positions.

15. Moreover, Dr. Haworth has reported that 40.8%^{^[14]} of applicants for MIT positions from Sam’s Club employees were women, while only 31.4%^{^[15]} of those promoted to MIT positions at Sam’s during 1996 through first quarter 2002 were women. At her deposition, she was asked if the MIT bid rate of 40.8% was compared to the selections during the period 1996-2002, whether the disparity would be statistically significant.^{^[16]} She first answered “I don’t know”^{^[17]}. When asked again, she said “I don't know, but one would calculate it.”^{^[18]} Finally, when asked a third time she answered “If you aggregated them all, I think they would be more than two standard deviations.”^{^[19]} Thus, according to Dr. Haworth’s own deposition testimony, there is a statistically significant female shortfall in actual promotions to MIT positions during 1996-2002 at Sam’s Club, compared to female availability based on applicants for the MIT program from Sam’s in January 2003.

Management Career Selection Incomplete

16. The MCS system is used by Wal-Mart to fill some openings in salary management jobs. Dr. Haworth concludes from her analysis of MCS data that there are no statistically significant selection decisions adverse to women across all the postings for each salaried job. This conclusion is misleading, because the MCS system cannot be considered as an unbiased, fair bidding system as described below, and covers only a small number of salary store management decisions.

17. The MCS system is rarely used to fill management positions below Store Manager. As Dr. Haworth reports (page 35 of her report, Declaration at 66:4-6), the MCS system was used to fill only 2% of Co-Manager positions, and less than 1% of Assistant Manager Positions. At her deposition, Dr. Haworth testified^{^[20]} that she could not say whether or not the small number promotions into Co-Manager or Assistant Manager positions found in the MCS data were a representative sample from those interested in these positions. Accordingly, no meaningful analysis of promotions into Co-Manager or Assistant Manager can rely on the MCS system. The only store salary management position for which the MCS system appears to have been used on any kind of regular basis is the Store Manager job.

18. The MCS appears to be used most of the time for filling Store Manager openings, but still the number of moves found in the MCS is about 400 fewer than the number found in the PeopleSoft data, according to Dr. Haworth’s calculations. (Report at 35; Declaration at 66:5). Dr. Haworth gives no explanation why there would 400 Store Manager openings filled outside of the MCS system. In Addition, as explained in my February 2003 report, the MCS system cannot be considered an unbiased bidding system due to the requirement that prior approval is necessary before an employee can bid. Moreover, Dr. Haworth fails to point out the fact that women promoted into Store Manager positions are disproportionately assigned to smaller stores than men, and hence earn less money.^{^[21]}

D. COMPENSATION

19. In my February 2003 report I presented several analyses of compensation of all hourly and salaried store management employees, based upon the statistical technique known as linear regression. Dr. Haworth presented an analysis of compensation for hourly employees, and for salaried store management employees restricted to Store Managers. She did not present any compensation analysis for Co-Managers and Assistant Managers in her report, or any other salaried employees, though she did perform them.^{^[22]}

20. Dr. Haworth also used linear regression analysis to do her compensation studies, but used a different model than I did. She claims that her studies “properly model the decision process”^{^[23]}. She cites deposition testimony, the Store Manager Survey, and certain statistical tests as the basis for her conclusions.

Compensation for Hourly Employees

21. Dr. Haworth incorrectly decided it is necessary to divide the hourly employees into about 7500^{^[24]} separate subunits and do separate regressions for each subunit. Moreover, she improperly includes certain explanatory variables in her regressions that are either tainted, or not considered by Wal-Mart in setting pay rates. Accordingly, her analysis is inaccurate and unreliable. The following paragraphs describe the defects in her analysis in more detail.

22. She ended up with such a large number of separate regressions by first dividing employees by store, then further dividing them by whether they held grocery or non-grocery jobs, then further dividing the non-grocery jobs into the six specialty divisions^{^[25]} and the remaining non-grocery jobs. In many stores, her analysis separates employees into eight subunits within a store^{^[26]}. There is no basis in Wal-Mart policy, or statistical justification for the extreme disaggregation of the data used by Dr. Haworth. The consequence of her disaggregation is to reduce the number of employees analyzed in each regression to a small group, sometimes as low as 20-30 employees.^{^[27]} In many cases there are so few employees in a subunit that the regression for this group could not be run.

Overly Disaggregated Analysis, Stores

23. Dr. Haworth ignores established, documented company wide policy controlling much of the compensation process, and contradicting her methodology. She cites Store Manager discretion^{^[28]} in setting pay rates as an important reason for her decision to do separate store regressions. In fact, Store Managers are constrained in the amount of discretion they have in setting pay rates. Wal-Mart’s company wide Field Associate Compensation Guidelines indicate several aspects of salary setting and job assignment which require approval at the District and/or Regional Manager level. The Compensation Guidelines state that “the Store Manager needs to have the flexibility to address ... differences”^{^[29]} among employees in setting pay rates. However, these Guidelines list several ways in which the Store Manager is constrained in their discretion:

a. “Exceptions to these guidelines will be reported every pay period in the Payroll Exception Report, which will roll up to the Distinct Manager and Regional Vice President.”^{^[30]}

b. “The People Group and your Regional People manager will act as consultants to ensure consistency in the program’s administration and to provide compensation standards for hiring, evaluating and awarding pay increases.”^{^[31]}

c. “A facility’s pay structure is based on local competitive pay rates of comparable jobs, and established in conjunction with the District Manager, Regional Vice President, and Regional People Manager.”^{^[32]}

d. In setting the starting rate for new hires, “... any increase above 6% of the Starting Rate requires District Manager or Specialty Group Regional Manager approval.”^{^[33]}

e. “... all associates’ pay levels should be reviewed and any pay inequities caused by the Start Rate adjustments should be identified and discussed with your District Manager.”^{^[34]}

f. A national pay structure specifies a $0.25 per hour gap between Start Rates in consecutive pay classes (i.e. pay class 1 to 2, pay class 2 to 3, etc.) ^{^[35]}

24. Moreover, store management employees frequently change stores, districts and even regions.^{^[36]} These personnel decisions made by Wal-Mart reflect control exerted above the Store Manager level, further indicating that stores are not isolated from each other. By doing separate regressions for every store subunit, Dr. Haworth fails to capture the effect of District, Regional, and company wide control over the compensation process.

Overly Disaggregated Analysis, Subunits within Store

25. Dr. Haworth claims each store contains too broad of a group of employees for making meaningful comparisons. So, she divides each store into as many as eight subunits for separate analysis (i.e. grocery, non-grocery, and the six specialty divisions. Her disaggregating of each store into sub-groups is not justified by any written Wal-Mart policy, nor did she conduct any statistical analyses to justify her assertions that the subunits she defines within a store have different “pay structures”.^{^[37]}

26. Dr. Haworth admitted at her deposition that Wal-Mart guidelines do not mention department as a factor to consider in setting pay rates.^{^[38]} Wal-Mart’s Field Associate Compensation Guidelines indicate how starting pay rates, and increases thereafter, are to be set for hourly jobs, and make no distinction between department or divisions. Thus, according to company wide Wal-Mart policy expressed in its Field Associate Compensation Guidelines, there is no separate “pay structure” for grocery, non-grocery, and each of the six specialty divisions.

27. There is a great deal of movement between departments^{^[39]} in a store, indicating that Dr. Haworth’s departmental subunits in a store do not have isolated decision making structures. Dr. Haworth testified at her deposition that it is ultimately the Store Manager’s decision to reassign employees between departments.^{^[40]}

28. Dr. Haworth’s disaggregating of stores into subunits within the store, and doing separate analysis for each sub-group, makes it impossible to identify important gender patterns that may occur in a store. For example, if men and women are disproportionately assigned to different departments, which are in separate sub-groups in Dr. Haworth’s analysis, then the pay rates for these men and women would never be included in the same regression, and therefore never compared. Store Manager Survey Improper and Unreliable

29. Dr. Haworth relies extensively on the manager survey^{^[41]} to justify her method of disaggregating employees and variable selection in regression model, despite admitting that the methodology of the survey violates accepted scientific standards, and was conducted in a manner contrary to her recommendations. Moreover, she acknowledged that having attorneys conduct the survey is considered to be biased and unreliable by courts and the scientific community. At her deposition she was asked, “Do you consider this survey as designed and as implemented to be a scientifically acceptable survey?”^{^[42]} She answered: “I don't know enough about the survey, and I'm also not a survey expert. I don't know enough about the survey and the way it was administered to be able to reach a judgment on whether it's a scientifically sound survey.”^{^[43]} However, Dr. Haworth and her staff were deeply involved in the design of the survey and made recommendations on how it should be implemented.^{^[44]} She knew the survey was conducted by attorneys in this litigation,^{^[45]} although she advised the lawyers for Wal-Mart that having the attorneys conduct surveys was not a good idea,^{^[46]} “[b]ecause typically it's difficult for an attorney to collect information in a neutral environment so that they truly get a neutral set of information back.”^{^[47]}

30. Dr. Haworth testified at her deposition that she was aware that the survey violated important principles in survey design listed in the “Reference Manual on Scientific Evidence”^{^[48]}, which she admitted is an authoritative treatise on scientific evidence in her deposition^{^[49]}, and cites in her report.^{^[50]}

31. In explaining how she developed her regression models, and determined which variables to include in those models, Dr. Haworth states:

“Second, we need to gain an understanding of the factors that the decision-makers rely upon when determining the pay rates for hourly employees. With the answers to these questions, the researcher is able to construct a statistical model that reflects the actual decision-making process as closely as possible.”^{^[51]}

She claims to rely on the store manager survey “to gain an understanding of the factors that the decision-makers rely upon when determining the pay rates for hourly employees.” However, Dr. Haworth fails to point out the most important single factor, cited by more store managers than any other factor as playing a role in determining starting hourly pay. The most frequently cited factor was: “The minimum pay established for the job classification by Wal-Mart’s pay guidelines.”^{^[52]} This same factor was also cited most often as being relied upon by store managers in setting promotional pay increases. Thus, company wide pay guidelines were found to be the most important factor that store managers rely upon in setting pay rates and pay increases. This fact is never pointed out in Dr. Haworth’s report^{^[53]}, and indeed, contradicts her interpretation that all store subunits operate with different pay structures and decision making processes. Although the store manager survey suffers from serious defects as described in the previous paragraphs, Dr. Haworth’s misinterpretation of the results of this survey undermine the justification she gives for her highly disaggregated models.

Improperly Excludes Hourly Department Heads

32. Dr. Haworth has inexplicably excluded all employees holding hourly Department Head^{^[54]} positions from her compensation regressions for hourly employees. Department Heads are among the highest paid hourly employees at Wal-Mart. Dr. Haworth’s exclusion of approximately 60,000 hourly employees from her analysis appears to be an error. She never mentioned this exclusion in her report or at her deposition, she never criticized my compensation analysis for including these hourly employees, and there is no reason Department Heads should be excluded. This apparent error in her analysis was discovered through examination of her backup materials that included her computer programs and the raw data files used as input for those programs.

33. Dr. Haworth constructed a variable ‘mgrsalever’, among others, for identifying employees she wished to exclude from her compensation regressions for hourly employees.^{^[55]} The programs used to run her hourly regression analyses include a statement which excludes any employee for whom the variable mgrsalever =1. Since Department Heads are identified by job code = 101, and every employee with job code = 101 has the value mgrsalever=1 in her raw data files, it follows that every employee with job code = 101 is excluded by her programs. It appears that her variable mgrsalever was designed to restrict her regressions to employees who “had never been salaried employees during their employment at Wal-Mart” specified at lines 4-5 of page 47 of her report. See Declaration at 101:17-18. For example, there are a small number of employees as of October 2001 (the date Dr. Haworth used for measuring pay rates in her analysis) who were currently in hourly jobs, but who were previously in salary positions such as Store Manager, Co-Manager, or Assistant Manager. Inexplicably, Dr. Haworth sought to exclude such employees, and set the variable mgrsalever =1 for these former salaried employees. Unlike Department Heads, other employees holding hourly supervisor jobs such as Support Manager (1050), CSM (510) and Lead (910) all have mgrsalever =0, and are included in Dr. Haworth’s regressions of hourly employees (unless excluded for some other reason).

34. If Dr. Haworth intended to include Department Heads in her hourly regressions, but excluded them, then her results are incorrect. On the other hand, if she intended to exclude them, her analysis is not probative. There are an average of about 30 Department Heads per store, and they account for approximately 13% of the hourly employees. As I pointed out in my February report^{^[56]}, women earned about $1800 less than men during 2001, among full-time Department Heads working over 45 weeks. Accordingly, presenting an analysis of compensation for hourly employees excluding Department Heads without giving any explanation is practically useless. After my rebuttal report was served and my second deposition taken, Dr. Haworth, in her Declaration at 101 n. 134, asserts that she has performed “alternative” regressions that purport to correct this error. I have not been provided any back-up material from which I could assess this assertion.

Arbitrarily Excludes Many Employees from Regressions

35. Aside from Dr. Haworth’s exclusion of Department Heads, she also excludes employees from her analysis based upon her use of arbitrary and unexplained restrictions. For example, she has excluded employees from her analysis of hourly employee pay rates if they have ever been demoted while working at Wal-Mart, or had ever been salaried employees during their employment at Wal-Mart.

Improperly Includes Tainted Variables

36. Dr. Haworth includes several tainted variables in her compensation analysis that have the effect of masking gender disparities in pay rates. Dr. Haworth has written in a law review article^{^[57]} that it is inappropriate to include variables in a model when the values of the variable itself may be influenced by employer discrimination. However, she has included several such variables in her models.

a. The gender distribution among departments is far from being a random distribution.^{^[58]} This uneven distribution originates at time of hire^{^[59]} as a result of Wal-Marts’ uneven gender assignment to initial department. The initial hire separation by gender is re-enforced by the job posting system, where women are significantly less likely to be promoted into the highly male departments, and significantly more likely to be promoted into the highly female departments^{^[60]}.

b. Starting pay rate is another tainted variable Dr. Haworth includes in her analysis. Table 15 on page 23 of my February 2003 report indicates that women are paid less than men at time of hire. That table shows women hired in 1996 were paid between 20 and 40 cents less per hour than men hired in the same job, on the average, for the jobs with the most hires. I have performed a more refined analysis of gender differences in starting pay rate, using Dr. Haworth’s data file provided with her backup materials. My analysis shows that initial pay rates for women are less than men’s, and this difference is statistically significant.^{^[61]}

c. Dr. Haworth includes among her explanatory factors the variable “whether or not someone has ever worked in a grocery division”. Again, this variable reflects uneven assignment of males to grocery departments, compared to women.^{^[62]}

d. Dr. Haworth includes the variable “whether someone was promoted in the past year” in her regressions, but gives no explanation or justification for including this variable. In fact, this is another improperly included tainted variable, since the promotion discrimination is an important issue in the case and my February 2003 report presents several results indicating that the promotion decisions at Wal-Mart have significant adverse impact on women.

Variables Included in Regressions Without Justification or Explanation

37. Aside from the tainted variables, there are several other variables that Dr. Haworth included in her regression model for which no explanation or justification is presented. These include:

Whether or not someone has changed stores at any time during their career;

Whether or not someone was hired as part-time or full-time;

Whether or not someone’s job code is one of the Sales Associate job codes^{^[63]};

Interaction term between Sales Associate and department;

Whether or not someone ever received a premium for working night shifts;

Whether or not someone had ever held secondary job responsibilities;

First pay group;

Division.

Incorrect Application of Chow Test

38. Dr. Haworth incorrectly applies statistical theory to perform her analysis. She refers to a statistical procedure known as the “Chow Test” to justify her decision to do separate regressions for every subunit of every store.^{^[64]} The test is named after its author, Gregory C. Chow. His original article is attached as Ex. 4. In fact, the statistical theory on which the Chow Test is based does not justify her conclusion that separate regressions must be run for every store sub-component.

a. The statistically significant results from the Chow Test would indicate only that there might be at least one variable that is different in two or more stores. It does not imply that every variable in every store has a different relationship to pay rate. In fact, the method she employed tells her nothing about which stores or which factors might be different.^{^[65]} Dr. Haworth’s regression implementation is equivalent to assuming that every factor in every store has a different relationship to pay rate. It could be, based on her variables, that only one store is different from other stores.

b. Dr. Haworth did not perform any Chow Tests for studying whether her store sub groupings into each of the specialty divisions are appropriate.^{^[66]} At her deposition she claimed to have done Chow tests comparing grocery and non-grocery sub-groups, but these are not mentioned in her report, and they were not included in her backup materials.

Fails to Re-Aggregate to Compute Overall Results

39. As explained above, Dr. Haworth separates employees into approximately 7500 subunits, does a separate regression for each subunit, but never reports any measure of disparity resulting from re-aggregating her subunits. This is contrary to principles she has espoused in other cases^{^[67]} and articles^{^[68]}, where she suggests computing an overall measure of disparity and its statistical significance when an analysis is done separately on independent subunits, as she has done with her regressions in this case. At her deposition, when asked “... did you ever aggregate all the individual results to see if overall there was a statistically significant pattern against women?”, she responded “I don't know how to do that - other than to show you the patterns that are here.”^{^[69]}

40. In fact, it is a straightforward statistical exercise to obtain an overall measure of disparity and corresponding measure of statistical significance by properly reaggregating the results computed separately from each of the sub-groups Dr. Haworth has created. Since the subunits include disjoint groups of employees, it’s possible to get a (weighted) average of the gender coefficients across the sub-groups. The calculation of the average^{^[70]} gender coefficient across Dr. Haworth’s sub-groups results in an average pay shortfall of $0.12 per hour for hourly employees. The t-value for this disparity is –7.22 indicating a statistically significant result, which would occur with less than 1 chance in 10 to the 11^th power by random fluctuation.

41. Thus, the reaggregated results computed from Dr. Haworth’s unjustified extreme disaggregating of hourly employees, and using her tainted and unexplained variables, still shows a statistically significant difference in average pay rate for men and women of about $0.12 per hour. This result is consistent with Dr. Haworth’s admission at her deposition that in each set of subunits in Wal-Mart Division 1, Sam’s Club, and SuperCenter grocery and non-grocery a majority of the subunits showed pay rate differences adverse to women, both for all regressions, and restricted to those that resulted in statistically significant gender coefficients.^{^[71]}

Effects of Her Methodology

42. Dr. Haworth’s extreme disaggregating of the employees makes her analysis unable to detect possible important gender differences. There are two types of situations where Dr. Haworth’s method will overlook or minimize important disparities.

a. There are many cases where men and women holding the same job are separated by their department, as well as other categorical factors Dr. Haworth includes in her models. Where men tend to be placed in departments with higher pay, then differences in pay between men and women in the same job would be attributed to the department variable, not the gender variable. The more variables that are included in the model, the more this situation will occur.

b. By doing separate regressions for each store subunit, Dr. Haworth will never detect the situation where men are paid more in one store subunit than women in another store subunit, even though they have identical values for her explanatory factors. Her models would attribute gender pay differences to different “pay structures” in different store subunits, even though no such differences exist.

E. COMPENSATION FOR SALARIED EMPLOYEES

43. In her report, Dr. Haworth presented an analysis of total compensation for Store Managers of Wal-Mart Discount and SuperCenters, and a separate analysis of Managers of Sam’s Clubs. She did not present any compensation analysis for Co-Managers and Assistant Managers.

44. Dr. Haworth compensation analysis for Store Managers is defective, because she includes tainted variables that mask the gender differences in earnings. As was noted earlier in this report, my analysis of promotions into Store Manager show that promoted women are disproportionately assigned to smaller stores. Therefore, in analyzing Store Manager compensation, "Square footage of the store", and "Number of employees at that store" are tainted variables, which mask compensation shortfalls of female Store Managers. Dr. Haworth also includes the variable "Store profit per square foot" without explanation or any justification why this would be a relevant and gender neutral factor.

I declare under penalty of perjury that the foregoing is true and correct.

Dated: July _____, 2003 _____________________________________

Richard Drogin

^{^[1]}See middle of page 24 of her report; Declaration at 47:17-19.

^{^[2]} See Table 19, pages 28-29 of Drogin, February 2003 report. Lateral moves are not counted in the PeopleSoft promotions in this study.

^{^[3]} Dr. Haworth was aware there was or might be a problem with the job posting data, since she had read my report, which indicated the job posting data was incomplete, and also attended my deposition at which I described the problem.

^{^[4]} Haworth deposition page 101, line 14.

^{^[5]} Haworth deposition, page 101, lines 15-24.

^{^[6]} The article appears in Volume 12, pages 352-369 of the Employee Relations law Journal, and has been designated in this litigation as Bates WMHO1234046-WMHO1234063. Quote appears on WMHO1234058, attached as Ex. 1.

^{^[7]} ‘% Women’ is the % women among active employees at year-end 2001, restricted to departments with at least 5000 employees, shown in Table 14 of my February 2003 report, ‘Diff’ and ’Z-Value’ are taken from Dr. Haworth’s table on page 22-23 of her report. See Declaration at 43-44.

^{^[8]} Dr. Haworth’s table on page 22 of her report shows that women received 62.6 fewer promotions into Department Head jobs in the Home Furnishing Department, resulting in a Z-value for this disparity of –3.55. The Home Furnishing Department had only 23% women among its employees at year-end 2001, but was not included in Table 1b above, because the number of employees in this department was below the 5000 level that I used for selecting departments. If Home Furnishings had been included in Table 1b, then the disparity, and pattern of under-promotion of women in predominately male departments would be more pronounced. Her declaration uses similar, although somewhat changed numbers, reflecting a shorter time frame. Declaration at 43-44.

^{^[9]} Page 25 of Haworth report. Declaration at 51:7-10.

^{^[10]} She also suggests the bid rate for Support Manager positions in the job posting data is consistent with the actual promotions into MIT positions, on page 25 of her report. However, in her deposition testimony, page 116, lines 6-10, she admitted that this comparison is not meaningful. Her Declaration does not include this suggestion.

^{^[11]} At her deposition, page 122, line 24 through page 123, line 16 Dr. Haworth indicated her opinion that the MIT selections were the “most qualified” applicants among those “interested and available”, and she had “no information” that affirmative action was being used to select women.

^{^[12]} Page 122 line 24 through page 123 line 6 of Haworth deposition..

^{^[13]} In my report, and at my deposition I explained that insufficient information was available at that time to evaluate the fairness of the January 2003 MIT postings.

^{^[14]} Page 30 of Dr. Haworth’s redlined report. Declaration at 53:21

^{^[15]} Page 89 of Dr. Haworth’s redlined report. Declaration at 165:6. My computations show only 25.4% of MIT promotions were women during 1997-1^st quarter 2002.

^{^[16]} Haworth deposition, page 151 lines 7-11.

^{^[17]} Page 151, line 12

^{^[18]} Page 151, line 21

^{^[19]} Page 152, lines 1-2

^{^[20]} Page 310, lines 11-16

^{^[21]} I examined whether women were disproportionately assigned to smaller stores. Promotions into Store Manager were divided up according to whether the target store was large or small, where ‘large’ stores were those with size 60,000 square feet or more. For each year, and for Sam’s and non-Sam’s stores, I determined the shortfall of women promoted into large stores compared to what is expected from their proportion among promotees. The overall Z-value for this analysis is –3.97, indicating a statistically significant pattern where women were moved into smaller stores.

^{^[22]} She admitted in her deposition that she had performed such analyses, see page 320, lines 14-18, and page 322 lines 14-18. Her backup materials indicate the results she found showed a pattern adverse to women.

^{^[23]} Haworth report, page 104. Declaration at 132:19.

^{^[24]} The total number of sub-units for which she attempted regressions is 7691, based on her supplemental results contained on the CD she turned over at her deposition. Also, see page 172 lines 15 through page 173 line 9 of her deposition.

^{^[25]} Haworth deposition, page 196, lines 10-25.

^{^[26]} For example, in SuperCenter stores, Dr. Haworth would separate employees into grocery, non-grocery, and 6 specialty division jobs.

^{^[27]} Haworth deposition, page 188-191.

^{^[28]} See page 41 of her report. Declaration at 92:16-93:2. On page 47, she says, “Because pay rates for hourly employees at Wal-mart and SC etc. are generally established by Store Mgrs etc.”

^{^[29]} Field Associate Compensation Guidelines, WMHO000676. The Guidelines are attached as Ex. 2.

^{^[30]} Field Associate Compensation Guidelines, WMHO000676.

^{^[31]} Field Associate Compensation Guidelines, WMHO000676.

^{^[32]} Field Associate Compensation Guidelines, WMHO000677.

^{^[33]} Field Associate Compensation Guidelines, WMHO000678. I found that during the period May 1999 through April 2000 (the one year period following the effective date of these Guidelines) approximately 90% of the stores had at least one instance where there was a new hire paid at least 6% above the starting rate in that store for the pay class into which the employee was hired. Moreover, I found that approximately 40% of all hires were initially paid at least 6% above the starting rate in that store for the pay class into which the employee was hired.

^{^[34]} Field Associate Compensation Guidelines, WMHO000687.

^{^[35]} WMHO205186

^{^[36]} See Table 16 and 17 of Drogin February 2003 report.

^{^[37]} The Chow tests she incorrectly claims justify her disaggregation of employees by store, were never applied to tests for differences between subunits within stores.

^{^[38]} Page 81, line 14 through page 82, line 8, and pages 201-203 of Haworth deposition.

^{^[39]} Haworth report, page 16, Declaration at 48:5-7.

^{^[40]} Haworth deposition, page 95, lines 14-17.

^{^[41]} See Haworth report at the top of page 42, and page 45, Declaration at 93:4-99:4, “The above description of the decision-making process makes it clear that there are multiple compensation structures and decision-making processes for hourly employees of Div 1 etc.” Also, page 40: (Declaration at 85:20-86:2) “... we must account for the factors used by decision makers at Wal-Mart to set salary levels and we must account for the different compensation decision-making processes found throughout the company.”

^{^[42]} Haworth deposition, page 288, lines 9-13.

^{^[43]} Haworth deposition, page 288, lines 21-25.

^{^[44]} Haworth deposition, pages 267-274.

^{^[45]} Haworth deposition, page 251, lines 7-25.

^{^[46]} Haworth deposition page 254, lines 3-7.

^{^[47]} Haworth deposition, page 254, lines 14-17.

^{^[48]} Reference Manual on Scientific Evidence, 2^nd ed, West Group, 2000, page 237-238, and Haworth deposition page 291, lines 1-16.

^{^[49]} Haworth deposition, page 290, lines 20-21.

^{^[50]} Haworth report, page 109, footnote 241. This citation is not included in her Declaration.

^{^[51]} Haworth report, page 41, Declaration at 92:12-15..

^{^[52]} Haworth report, Appendix C-7, (appendix C-16 to her Declaration) and Haworth deposition page 276, lines 9-17. The second most frequently cited factor that Store Managers said they took into account in determining starting pay rates was the “Starting rate in the department in the store at the time the offer is made.” This factor is ‘nonsensical’ because there is no starting pay for a department. The Field Compensation Guidelines indicate there is a starting pay rate for each pay class, regardless of the department. Dr. Haworth agreed that there is no starting pay for a department at her deposition (see page 202 line 10 to page 203 line 3, and page 295 line 19 through page 296 line 8).

^{^[53]} On pages 42-44 of her report,(Declaration 93-98) Dr. Haworth lists the factors included in the manager survey, but does not give the percentages of managers who said they relied on each factor, and she could not give any rationale for how the factors were ordered on the 3 page list, as stated on page 275 of her deposition. The most important factor, “The minimum pay established for the job classification by Wal-Mart’s pay guidelines” is listed last on page 44 of her report, and in her Declaration at 98.

^{^[54]} Department Heads are designated by job code = 101.

^{^[55]} These programs are contained on a CD provided at her April 21, 2003 deposition.

^{^[56]} Drogin report, Table 10.

^{^[57]} Notre Dame Lawyer, vol 54:633, on page 656. Ex. 3. Also, see the article “Advanced Statistical Techniques – Compensation Analysis”, page 8, 2^nd paragraph from bottom. This page is designated as WMHO1234022 in this litigation. Ex. 4

^{^[58]} For example, see Table 14 on page 21 of my February report.

^{^[59]} I analyzed the assignment of new hires to departments, and found that women were disproportionately assigned at hire into the 10 departments with the highest percent female, and found the disparity to be highly significant (Z=125.59). This analysis was conducted on all hires at Wal-Mart, using the department at the time of hire, store, year of hire, starting status (pt or ft).

^{^[60]} See Tables 1a and 1b above.

^{^[61]} I compared the starting pay rates of men and women hired into hourly jobs, in the same store, in the same year, in the same starting pay group, and having the same first status (pt/ft), based on Dr. Haworth’s raw data file of hourly employees active or on leave as of October 2001. The disparity for this comparison has a Z-value of -71.63, indicating a high degree of statistical significance for the shortfall in female starting pay rates.

^{^[62]} I compared the percent of men to the percent of women who have ever held a grocery job among those included in Dr. Haworth’s raw data file of hourly employees active or on leave as of October 2001. The analysis controls for year of hire, and first store, and results in a Z-value of -71.36, indicating a high degree of statistical significance for the pattern that the percent of women who were “ever grocery” is less than the percent for men, based on Dr. Haworth’s raw data.

^{^[63]} It should be noted that indicators for each Job code are already included as separate variables.

^{^[64]} Page 47 of Haworth report, “A statistical test called a ‘Chow’ test allows us to determine whether there are statistically significant differences between stores with respect to their compensation structures. If the structural differences between stores are statistically significant, then there is also a statistical justification for conducting a separate regression analysis for each store.”

^{^[65]} Haworth deposition, page 182, lines 15-22.

^{^[66]} At her deposition she could not recall doing any Chow test for testing whether her models differed in any way among specialty departments, and was not certain whether she had done Chow tests, which would have led her to separate grocery and non-grocery. See page 180 lines 25 – page 181 line 4, and page 179 lines 8-19. After her deposition, I was provided with a new disk of data that purported to include grocery/non-grocery chow tests. The date of the output for this analysis indicates that it was done after her deposition.

^{^[67]} For example, see “Affidavit of Joan Haworth”, sworn on June 14, 1994, in Thomas v. Christopher, on page 3. Ex. 6 Also, see her report “Statistical and Economic Characteristics of Ingles Markets and Workforce “ dated April 12, 1998, on page 8, and designated as WMHO1227076 in this litigation. Ex. 7.

^{^[68]} See page 8 of Dr. Haworth’s article “Economics and Statistics in the Employment Environment”, designated as WMHO1234043 in this litigation. Ex. 8.

^{^[69]} Haworth deposition, page 231, lines 19-24.

^{^[70]} The average gender coefficient is computed by taking the weighted average of the gender coefficients Dr. Haworth found in her individual sub-groups weighted by the number of women in the sub-group. This corresponds to the average dollars per hour women are paid less than men, after controlling for store, all the independent variables Dr. Haworth uses, and also all possible interactions between store and her independent variables. The calculation is made from the backup data files provided by Dr. Haworth, using her results for the model, which do not include starting pay rate as a variable.

^{^[71]} Haworth deposition, page 229 lines 10 through page 231 line 5.

NORTHERN DISTRICT OF CALIFORNIA

Defendant

I, Dr. Richard Drogin declare:

A. INTRODUCTION

Effects of Her Methodology