222 S.W. Columbia St., Suite 1600 Portland, OR 97201 Phone: 503-222-6060 www.econw.com
This page intentionally left blank.
CONTACT INFORMATION This report was prepared by Dr. Bryce Ward of ECONorthwest, which is solely responsible for its content. ECONorthwest specializes in economics, planning, and finance. Founded in 1974, were one of the oldest independent economic consulting firms in the Pacific Northwest. ECONorthwest has extensive experience applying rigorous analytical methods to examine the benefits, costs, and other economic effects of environmental and natural resource topics for a diverse array of public and private clients throughout the United States and across the globe. For more information about ECONorthwest, visit our website at http://www.econw.com. For more information about this report, please contact: Bryce Ward ECONorthwest 222 SW Columbia Street Portland, OR 97201 503-222-6060
This page intentionally left blank. ECONorthwest The Economic Costs of Medical Errors in Oregon iii TABLE OF CONTENTS CONTACT INFORMATION ....................................................................................................... I TABLE OF CONTENTS ......................................................................................................... III I. BACKGROUND AND SUMMARY OF FINDINGS ...................................................................... 1 II. DATA AND METHODS .................................................................................................... 3 A. Data .................................................................................................................................... 3 B. Methods .............................................................................................................................. 4 III. RESULTS ...................................................................................................................... 6 IV. EXPLANATION ............................................................................................................. 12 A. Comparison to Capitol Hill Site ......................................................................................... 12 B. TFK difficulties obtaining qualified laborers ..................................................................... 14 C. Evidence from applications and EEOC responses .......................................................... 15 D. Evidence from jobs employing same management in other locations ............................. 17 V. CONCLUSION .............................................................................................................. 20 APPENDIX A SELECTED APPLICATION AND INTERVIEW MATERIALS ................................... 21 ECONorthwest Evaluation of Employment Outcomes at TFK J/V 1 I. BACKGROUND AND SUMMARY OF FINDINGS The Traylor -Frontier-Kemper Joint Venture (TFK) 1 faces accusations that its employment practices discriminate against minorities, particularly African Americans. A statistical analysis by Dr. Nayak Polissar and Moni Neradilek prepared for Sound Transit (hereafter the Polissar and Neradilek report) documented racial disparities in employment outcomes at TFK. 2 Attorneys representing TFK requested that ECONorthwest (ECONW) review the Polissar and Neradilek report and conduct original statistical analyses of employment practices at TFK. The data available do not support statistical conclusions about potential discrimination at TFKs Sound Transit project. While Polissar and Neradilek document statistical disparities in outcomes, their report contains several flaws that render it incomplete and unreliable. Specifically: (1) The data provided by Sound Transit and used by Polissar and Neradilek contain numerous inconsistencies that may render conclusions based on these data inaccurate and unreliable. The data analyzed by Polissar and Neradilek differ substantially from the data we examined even though both data sets originated at Sound Transit. Some individuals in the data we examined do not appear in the Polissar and Neradilek data, and some individuals in the data examined by Polissar and Neradilek do not appear in the data we examined. For the individuals included in both datasets, key information like ethnicity, hire date, termination date, and hours worked is not identical in both observations. That is, a persons ethnicity, termination date, etc. in one data set does not always match the ethnicity, termination date etc. for the person with the same name in the other data set. Given the relatively small number of observations in the data, these inconsistencies potentially render any analysis based on these data inaccurate and unreliable. (2) Polissar and Neradilek omit information on union membership, worker classification, and reason for departure from their analysis, and this omitted information matters. Polissar and Neradilek ignore differences in outcomes for workers outside the laborers union, they ignore differences in outcomes across groups within the laborers union (e.g., apprentice, group 3, group 5), and they ignore the reasons for workers departures. Each of these factors provide important context for understanding employment outcomes at TFK. By excluding this information, Polissar and Neradilek miss the fact that disparities in outcomes between ethnic groups are limited to two subcategories of workers from a single union who were disproportionately turned around or dismissed for reasons related to performance. TFK has ample documentation confirming that it struggled to obtain qualified workers in the affected categories regardless of ethnicity and the data are consistent with their claims.
1 TFK is a joint venture formed to construct light-rail tunnels between the University of Washington and Capitol Hill for Sound Transit in Seattle, WA. 2 Polissar, N. and M. Neradilek. Ethnicity and Employment at Sound Transits UW and Capitol Hill 2 Polissar, N. and M. Neradilek. Ethnicity and Employment at Sound Transits UW and Capitol Hill Worksites The Mountain-Whisper-Light Statistics, April 18, 2012. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 2 (3) Polissar and Neradileks analysis comparing TFKs outcomes to outcomes at different Sound Transit project does not support meaningful conclusions because it rests on faulty assumptions. Polissar and Neradilek compare TFKs employment outcomes to outcomes from a similar Sound Transit project (Capitol Hill); however, for this analysis to prove useful, one must assume that both sites drew workers from the same labor pool, to do the same work, using a random process. These assumptions are invalid, and, as such, this analysis does not provide useful information. (4) Polissar and Neradilek do not investigate the possibility that differences in outcomes across ethnic groups reflect differences in the qualifications and skills among those dispatched. As Polissar and Neradilek state in their report, the only potentially causative factor studied is ethnicity. Qualifications and other factors were not considered. Clearly, though, qualifications and other factors matter. An analysis of information contained in a mix of employment applications, interview notes, and resumes for a subset of the laborers dispatched to TFK indicates that the laborers who were turned around or dismissed due to inability to perform were substantially less likely to provide clear evidence of prior experience on large tunneling projects than laborers who were hired and not dismissed for reasons related to job performance. While the information relied upon in this analysis does not include all TFK laborers and may not provide a complete and accurate description of the qualifications for the individuals examined, this analysis certainly suggests that differences in qualifications were important. As such, to assert that the observed disparities are the result of discrimination requires one make the baseless assumption that no differences in qualifications or other relevant factors exist across the relevant individuals. In the remainder of this report, we describe the results of our analysis and each of these key findings in more detail. We find that statistically significant disparities in outcomes between African Americans and other ethnic groups are limited to a subset of occupations within a single union laborers group 6 or higher (i.e., tunnel miners). Quantitative and qualitative evidence indicates that TFK had significant difficulties finding qualified workers in these groups, regardless of ethnicity, and that the tunnel miners dismissed were dismissed for reasons related to performance or qualification. This supports TFKs assertion that worker qualifications explain observed differences in outcomes. We also demonstrate that, at previous jobs in other cities that employed the same TFK supervisory team, African American outcomes do not statistically differ from outcomes for other ethnic groups. This also supports TFKs assertion that worker qualifications explain observed differences in outcomes. Combined, these facts suggest that it is plausible that unobserved differences in worker qualifications explain the observed differences; however, without more complete data on the qualifications and abilities of the workers at issue, we cannot definitively explain the observed pattern of results. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 3 II. DATA AND METHODS Statistical analyses of employment outcomes typically address three main questions: (1) Do we observe meaningful differences in outcomes across groups? (2) Do observed differences stem from group status? That is, does group status cause differences in outcomes because group members face discrimination or do differences across groups stem from other, potentially unobserved, factors like qualifications or experience that may be correlated with group status? Stated in the language of statistics, are estimated differences inaccurate or biased? (3) Assuming observed differences are unbiased, are observed differences likely due to chance (i.e., are differences statistically significant)? A. Data To the extent possible, we address these questions using data provided to us by TFK representatives. Specifically, we examine data that includes the following: Ethnicity Union Membership Worker classification (as identified on dispatch slip) Dispatch date Total hours worked Separation date (if any) Reason for termination 3
In the analyses described below, we rely on the information we were provided specifically, the data in the most recent file we were provided -- Graphs as of 52912.xlsb; however, we also examined the data used by Polissar and Neradilek. 4 As we understand it, all datasets ultimately rely on data maintained by Sound Transit. 5
It is very important to note that the data available contain several significant potentially fatal flaws. The data examined by Polissar and Neradilek do not match the data we examined. While we do not understand the source of these differences at this time, the differences we observe are very troubling. First, the list of people named in the different datasets are not identical. Setting aside the dozens of individuals whose names have slight differences in spelling across the data sets (which weve assumed are typos and not separate individuals), the data we received includes seven individuals who are not included in the Polissar and Neradilek data even though their orientation dates indicate they were dispatched during the period
3 Specifically, use files provided by TFK labeled Graphs as of 52912.xlsb (specifically the data in the input graphs tab), TFKJV Weekly Employee hours for 2010 & 2011.xlsx, and U220-W.xisx. 4 The files we reviewed from Polissar and Neradilek were labeled U220 04 05 12 Analysis Hiring and Termination Data Revised.xls and U230 03 19 12 Analysis Hiring and Termination Data Revised.xls 5 Information from Jeremy Saperia ECONorthwest Evaluation of Employment Outcomes at TFK J/V 4 covered by the Polissar and Neradilek data. Similarly, the data examined by Polissar and Neradilek include five individuals whose names are not included in the data we received. As such, 8 percent of the total number of laborers across both datasets are not included in one data set or the other. Second, among the 137 laborers in both data sets, 13 (or nearly 10% of the total number of observations) have a different ethnicity code. For instance, 2 individuals are coded as African American in the data set we received, but are coded as Caucasian in the Polissar and Neradilek data, and one individual coded as Caucasian in our data is coded as African American in their data. 6
Third, dispatch/orientation dates, termination dates, and hours worked differ for several individuals in both data sets. For instance, 8 individuals have dispatch/orientation dates that differ by more than two weeks across both data sets. Two individuals show up as terminated in the Polissar and Neradilek data, but not terminated in our data, and two different individuals have termination dates more than 100 days apart between the two data sets. Finally, total hours worked are inconsistent across data sets. 7 In particular, hours differ by relatively significant amounts for most individuals who worked hours in 2011. The existence of such differences across data intended to describe the outcomes for the same set of individuals cast doubt on the reliability of any statistical analysis based on these data. Given the relatively small number of individuals included, data errors reflecting relatively few individuals could significantly affect the magnitude of observed differences and the statistical significance (or the reliability) of the results. Until these data issues can be satisfactorily resolved, any statistical analysis should be interpreted with extreme caution. B. Methods Setting aside concerns about the reliability of the data, our statistical analyses rely on methods similar to those used by Polissar and Neradilek. Because TFKs project is ongoing, we have not observed final outcomes for each worker. Thus we cannot compare final outcomes for all workers because we have not observed them. Statisticians refer to such data as censored, and they use a particular form of statistical analysis known as survival analysis (also know as time to event or duration analysis) when examining such data. 8
6 Even within a single excel file, some individuals have multiple ethnicities, names, etc. For instance, the U220-W.xis file contains tabs for which describe hours worked per week for each worker in each year, 2 individuals change ethnicity code across years, a few change name spelling, etc.. This data set also has six individuals who are listed as laborers in the Polissar and Neradilek data and in Graphs as of 52912.xlsb, listed as operators. 7 This description compares the Polissar and Neradilek hours to the hours reported in TFKJV Weekly Employee hours for 2010 & 2011.xlsx 8 For additional details regarding survival analysis see UCLA Academic Technology Services, Statistical Consulting Group Survival Analysis with Stata http://www.ats.ucla.edu/stat/stata/seminars/stata_survival/default.htm (accessed May 21, 2012), Kennedy, P., 2008. A Guide to Econometrics 6E. Malden, MA: Blackwell Publishing. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 5 Unfortunately, though, one cannot reliably address the fundamental question of interest using the available data. We do not have complete, reliable data on individual qualifications or skills. We only have information gleaned from a mix of employment applications, interview notes, and resumes for a subset of TFK workers. As such, we can only partially assess question (2) do observed differences reflect group status or do they reflect other attributes that may be correlated with group status in the data (i.e., qualifications and skills). We cannot definitively explain the source of any observed differences. To assert that observed differences reflect discrimination, one must assume that no differences exist in qualifications or skills across groups in the sample of individuals we actually observe. We are aware of no evidence that would support such an assumption, and the evidence available appears to directly contradict it. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 6 III. RESULTS Before describing outcomes by ethnic group, we examine some general descriptive facts about TFK workers. Table 1 presents some general facts about employment outcomes for each union. In this table and many of the others in this section, we present raw counts (as opposed to percentages). We do this to make very clear the relatively small sample sizes that underlie this analysis. Through May 25, 2012, 282 workers had been dispatched to the TFK project. Approximately 57 percent came from the laborers union, another 26 percent came from the operators union, and the remainder were split between the carpenters and the cement masons unions. A small number of people (16 mostly laborers) were dispatched on two separate occasions. 9 Thus, 264 different individuals were dispatched to the TFK site. The median job tenure is longest for operators (394 days) and shortest for carpenters (93 days) and laborers (188 days). In part, the brevity of laborers tenure reflects their later median start date. Table 1. General Facts About Employment Outcomes for Each Union Number of dispatches Number dispatched individuals Number dispatched 2 separate times Median number of days on job Median start date Carpenters 34 33 1 93 19-Nov-10 Cement Masons 12 10 2 260 13-Jul-10 Laborers 160 149 11 188 20-Jun-11 Operators 74 72 2 394 18-Mar-11 Total 280 264 16 222 11-Apr-11 Source: ECONorthwest analysis of data provided by TFK Table 2 describes the ethnic composition of workers dispatched from each union. Of the minorities dispatched to TFK, most (75%) came from the laborers union.
9 Some of these reflect people laid-off after temporary work ended who were called back. Some represent people who were turnaround that were dispatched again. For instance, TFK turned around some laborers who were qualified for cross-path work, but not for the tunnel boring machine (TBM) work for which they were hiring. Email from Michael Krulc June 27, 2012. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 7 Table 2. Ethnic Composition of Workers Dispatched from Each Union Carpenter Cement mason Laborer Operators Total African American 1 1 29 3 34 Asian 2 0 2 0 4 Caucasian 25 4 68 59 156 Hawaiian/Pacific Isla 0 1 0 1 2 Hispanic 1 2 31 3 37 Native American 1 1 5 2 9 Other 3 1 14 4 22 Total 33 10 149 72 264 Source: ECONorthwest analysis of data provided by TFK TFK data describe the reason for each job separation. Separations are grouped into one of seven categories: accepted other work, medical reasons, end of temporary work/reduction in force, failed to report/absences, falsification/no dispatch/insubordination/violation of safety rules, inability to perform, and turnaround. Table 3 describes the reason workers left TFK by union. Table 3. Reason Workers Left TFK, by Union Carpenter Cement Mason Laborer Operators Total Accepted Other Work 2 0 6 7 15 Medical 0 0 4 0 4 End Temp Work / Reduction of Force 26 9 17 10 62 Failed To Report, absences 1 0 3 1 5 Falsification; No Dispatch; Insubordination; violation safety rules 1 0 4 3 8 Inability to Perform 1 0 17 1 19 Turnaround 0 0 19 0 19 Total 31 9 70 22 132 Source: ECONorthwest analysis of data provided by TFK There are clear differences across unions in the reasons for departure. One hundred percent of the 19 workers who were turned around came from the laborers union. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 8 Similarly, 89 percent of the workers dismissed due to inability to perform came from the laborers union. Overall, 84 percent of workers dismissed due to worker performance or abilities (the bottom four categories in the table) came from the laborers union. Given that laborers make up only 57 percent of total dispatches, if terminations were random or proportional to dispatches, we should expect to observe 57 percent of dismissals in these bottom four categories. Instead, the proportion is nearly 50% higher than expected. This indicates that workers from the laborers union had significantly higher chances of being dismissed for one of these performance related reasons. 10
Table 4 breaks down the reasons for separation by ethnicity and union. The top panel presents the aggregate results for the non-laborer unions. The bottom panel presents the results for the laborers union. Among non-laborers, no one is turned-around and only eight are dismissed for inability to perform, insubordination, etc. A simple univariate survival analysis produces an expected number of events (separations) for each ethnicity taking account of how long workers have been on the job and compares the expected number to the actual number observed in the data. 11 For workers in non-laborer unions, if African Americans and Caucasians have the same probability of separation (for any reason), then we would expect African Americans to experience 2.42 separations and Caucasians to experience 46.58 separations. In actuality, African Americans experience 3 separations and Caucasians experience 46 separations. This difference between the actual and expected number of separations is not statistically significant. The p-value (roughly the odds that such a result would be observed by chance) is 0.70, far higher than the 0.05 level commonly used to denote statistical significance. This high p-value indicates that that small difference is more likely due to chance than due to systematic differences across groups. The bottom line here is this outside of the laborers union there are trivial and not statistically significant differences in employment outcome based on ethnicity. 12 We turn next to an analysis of outcome differences within the laborers union.
10 The p-value from a survival analysis, log-rank test comparing the chances of observing dismissals for these reasons across unions is 0.0002. 11 For each of the survival analyses/log-rank tests we present, for simplicity, we assume each event is independent. That is, for the individuals we observe twice, we treat each experience as a separate observation. The pattern of results we describe is unaffected by performing more sophisticated analyses that account for some individuals with multiple events (e.g., Anderson-Gill Cox models or conditional risk set models). 12 Including all ethnicities does not change this conclusion. Similarly, limiting the events to only dismissals for performance related reasons does not change this conclusion. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 9 Table 4. Reasons for Separation, by Ethnicity and Union African America n Asian Caucasian Hawaiian / Hispanic Native America n Other Total Non-Laborer Unions Accepted Other Work 0 0 9 0 0 0 0 9 Medical 0 0 0 0 0 0 0 0 End Temp Work / Reduction of Force 2 2 33 2 3 1 2 45 Failed To Report, absences 0 0 2 0 0 0 0 2 Falsification; No Dispatch; Insubordination; violation safety rules 0 0 3 0 0 1 0 4 Inability to Perform 1 0 1 0 0 0 0 2 Turnaround 0 0 0 0 0 0 0 0 Total Non-Laborer Unions 3 2 48 2 3 2 2 62 Laborers Accepted Other Work 1 0 4 0 0 1 0 6 Medical 3 0 1 0 0 0 0 4 End Temp Work / Reduction of Force 4 0 7 0 5 1 0 17 Failed To Report, absences 1 0 2 0 0 0 0 3 Falsification; No Dispatch; Insubordination; violation safety rules 0 0 1 0 2 0 1 4 Inability to Perform 6 0 7 0 2 0 2 17 Turnaround 6 0 5 0 6 0 2 19 Total Laborers 21 0 27 0 15 2 5 70 Source: ECONorthwest analysis of data provided by TFK The bottom panel of Table 4 presents the outcomes for the laborers union broken down by ethnicity. For each of the large ethnic groups (African Americans, Caucasians, and Hispanics) from the laborers union, a large share of people who cease work at TFK were either turned around or dismissed due to inability to perform. Of the 21 African Americans who left work at the TFK site, 6 were turnarounds and 6 were dismissed for inability to perform. Of the 27 Caucasians who left the TFK site, 5 were turnarounds and 7 were dismissed due to inability to perform. Of the 15 Hispanics who left the TFK site, 6 were turnarounds and 2 were dismissed due to inability to perform. While each of the major ethnic groups experienced a significant number of separations including a significant number of turnarounds and dismissals for inability to perform, African Americans experienced a disproportionate share of separations. Consistent with the findings in Polissar and Neradilek, the differences between African Americans and ECONorthwest Evaluation of Employment Outcomes at TFK J/V 10 the other ethnic groups within the laborers union are greater than we would expect due to chance alone. The results from a survival analysis similar to the one described above for the non-laborer unions indicate that, if African Americans and Caucasians experienced separations at the same rate, we would expect to observe 11.53 separations for African Americans and 36.47 separations for Caucasians. In actuality, we observe 21 separations for African Americans and 27 separations for Caucasians. These differences are statistically significant (p-value <0.01). The disparity in separation patterns does not simply exist across unions; it also exists across worker classifications within the laborers union. The data we obtained also contain information about each workers group (e.g., Laborer group 5 or TW group 6). In Table 5, we aggregate laborers into four categories apprentices, group 3 and group 4, group 5, and TW group 6 or higher and present the frequency of separations for workers in each. The vast majority (89%) of all dispatched workers dismissed due to individual performance (the bottom four categories in the table) fall in either group 5 or TW group 6+ -- the high qualifications groups. Among the dispatched group 5 or group 6+ laborers, over one-third were dismissed due to individual performance. Table 5. Frequency of Separations for Workers, by Laborer Category Apprentice Group 3 & 4 Group 5 Group 6+ Total Accepted Other Work 1 1 3 1 6 Medical 2 0 1 1 4 End Temp Work / Reduction of Force 0 11 3 3 17 Failed To Report, absences 0 0 1 2 3 Falsification; No Dispatch; Insubordination; violation safety rules 0 0 1 2 3 Inability to Perform 3 1 7 6 17 Turnaround 1 1 5 12 19 Total 7 14 21 27 69 Source: ECONorthwest analysis of data provided by TFK Disparities in employment performance across ethnic groups are also concentrated within the high qualification labor groupings. Table 6 presents the expected and actual number of separation events, derived from a survival analysis, for African Americans and Caucasians for each of the four groups of laborers presented in Table 5. For only one group, TW group 6+, do we observe large differences that reach the 0.05 percent level of statistical significance. Differences between expected and actual outcomes also exist for group 5; however, these differences are not significant at the commonly used 0.05 level (p=0.12). Differences in the likelihood of separation within group 3 and group 4 and ECONorthwest Evaluation of Employment Outcomes at TFK J/V 11 among apprentices are small and not close to statistically significant at the commonly used 0.05 percent level. Thus, the statistical significance in adverse employment outcomes for African Americans appears exclusively in the group 6 subset of a subset of workers. 13
Table 6. Results of Survival Analysis by Laborer Worker Group Apprentice Group 3 & 4 Group 5 Group 6+ Expected Events African Americans 3.98 4.3 1.32 4.42 Actual Events African Americans 4 4 3 10 Expected Events -- Caucasians 2.02 4.7 12.68 14.58 Actual Events -- Caucasians 2 5 11 9 P-value 0.99 0.84 0.12 0.001 Source: ECONorthwest log-rank test of data provided by TFK In sum, the simple analyses described above indicate: (a) relative to Caucasians (and other ethnic groups), African Americans are more likely to have experienced any separation or a separation related to individual performance and (b) differences in African Americans employment outcomes are not ubiquitous instead, they are concentrated within a single union (laborers) and within specific job classifications within that union (TW group 6 and, to a lesser extent, group 5). Polissar and Neradilek describe (a) but they do not describe (b), and (b) is important. A concentrated disparity indicates that the cause of the disparity is not common to the entire TFK project. Rather, the source of the disparity is concentrated in the parts of the project that rely on group 6+ (and perhaps group 5) laborers.
13 Different combinations of groups do not change this fundamental finding. E.g., grouping apprentices with group 3 and 4 does not yield large or statistically significant differences. Similarly, grouping laborers apprentices with groups 3 and 4, and with all non-laborers does not yield statistically significant differences. Even an analysis that included every one but group 6+ miners does not yield differences that are statistically significant at the commonly used 0.05 level. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 12 IV. EXPLANATION The results described above indicate that, for a particular subset of a subset of workers (group5/6 miners from the laborers union), a disparity exists between African American outcomes and Caucasian outcomes. However, as described in section III.A above, due to uncertainty about the quality of the raw data, we cannot be certain that this statistical disparity is real. Furthermore, even if the pattern of results above is accurate, these raw statistics do not explain the observed pattern. In the absence of explanation, we cannot interpret these differences. In this section, we investigate potential explanations for the observed results using the limited data available to us. As we understand it, the sides in this matter offer two competing explanations for the observed differences. One side, the dismissed employees, argues that the pattern described above indicates discrimination by TFK against African Americans. The other side, TFK, argues that the pattern reflects larger problems they had with the quality of group 5/6 workers dispatched by the laborers union and that the dispatched African Americans happened to be disproportionately unqualified in that subset. To distinguish among these competing stories we would ideally obtain extensive data on each dispatched workers qualifications and abilities to perform the work required on this specific project. For instance, we would like data that described each workers experience with the tasks required in tunnel mining, their experience working underground, and performance evaluations from previous jobs that describe work ethic, attitude, etc. We do not have these data (we only have partial information obtained from employment applications, interview notes, and resumes for a subset of TFK workers). 14
As such, we cannot directly distinguish among the competing stories by evaluating the outcomes of similarly qualified workers of different ethnicities (e.g., through a multivariate regression or some other form of multi-variable analysis). A. Comparison to Capitol Hill Site Polissar and Neradilek attempt to partially get around this limitation by comparing the employment outcomes at TFKs University of Washington (UW) site to the outcomes at the Capitol Hill (CH) site where Jay Dee/Coluccio/Michels Joint Venture (JCM) is the contractor. They include this comparison because the CH site draws workers from the same union hall as the UW site, and, according to Sound Transit staff, the work is similar for the relevant employees, but there is a different contractor. 15 This approach allows one to distinguish among the competing stories if one assumes that both sites drew workers from the same labor pool, to do the same work, using a random process. Under these assumptions, the quality of workers dispatched to each site for each ethnic group should be the same. 16 As such, given a sufficiently large sample, the implicit theory
14 We describe the available information and analyze it in more detail below. 15 Polissar and Neradilek p.2 16 Technically, the underlying distribution of quality (e.g., the average quality) should be the same for al ethnic groups, given a sufficiently large sample. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 13 appears to be that any differences between worker outcomes at the two sites reflect differences in the contractor and not differences in the quality of workers dispatched. 17
While Polissar and Neradilek find that African Americans have a poorer work experience at the UW site, the assumptions outlined above are not valid. As such, it is incorrect to conclude, based on this analysis, that contractor differences (e.g., discrimination) explain the observed pattern. To their credit, Polissar and Neradilek explicitly draw no such conclusions in their report. Comparing the UW site to the CH site does not support meaningful conclusions for several reasons. First, as we understand it, the process via which workers are matched to jobs is not random. Contractors have the ability to give preference to workers they have worked with previously (rehires) and call out specific workers by name. One of the joint venture partners comprising JCM, Coluccio, is local, but TFK is not (nor are Traylor or Frontier-Kemper). As such, JCM have much more knowledge about the local labor pool and potentially were able to select a greater percentage of their workforce directly. Second, as we understand it, while the CH and UW jobs are similar, they are not identical. The UW job is much larger. Through the periods examined in the Polissar and Neradilek report, the UW job had employed more than twice as many laborers as the CH job. The UW job is also more challenging. For example, the pressures to which tunnel workers could be exposed exceeded 5 bar. In contrast, pressures on the CH job only reached 1.7 bar. 18 Higher pressures meant that TFK required more workers with specialized experience and training. The data clearly suggest that these (and perhaps other) differences matter. Worker outcomes differ across sites for all ethnicities, not just African Americans. For instance, zero workers were turned around at the CH site. No African Americans. No Caucasians. No Hispanics. In contrast, 19 workers were turned around at the UW site, including 6 African Americans, 5 Caucasians, and 6 Hispanics. Similarly, very few workers appear to have been dismissed for inability to perform after only a short period on the job at the CH site. While we do not have data that describe the reason for separation at the CH site, at the UW site 15 of the 17 laborers dismissed for inability to perform were dismissed within 40 days. This suggests that TFK put some marginal laborers on the job and dismissed them relatively quickly once their quality was revealed. 19 In total, 69 percent of all laborer separations at the UW site occurred within 40 days of orientation. In contrast, only 5 out of 16 (or 31 percent) of separations occurred within 40 days of dispatch at the CH site. 20
17 We note that Polissar and Neradilek do not state this logic explicitly. 18 Interview with Michael Krulc, April 25, 2012. 19 See various TFK EEOC responses to request for information for EEOC Charge No. #551-2011-01391, EEOC Charge No. #551-2011-00415 Amended, EEOC Charge No. #846-2012-10533, EEOC Charge No. #551-2011- 01373 20 Expanding the period to 70 days or shortening to 30 does not eliminate the disparity. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 14 The disparities between the UW and CH sites, even among Caucasians, suggests that either TFK received less qualified workers from the union hall (perhaps because it lacked the local advantage described above), TFK set higher standards for their workers (perhaps due to the challenges on this job described above), or both. B. TFK difficulties obtaining qualified laborers Regardless of the specific reason for the difference between the UW and CH sites, the data strongly support TFK's repeated claim that they struggled to find qualified miners (laborers in groups 5 and higher). As noted in section IV, TFK turned around or dismissed a substantial share of the workers dispatched by the laborers union and the vast majority of turnarounds and separations due to the inability to perform were among laborers in groups 5 and higher. TFK's difficulties obtaining miners who met their standards are well documented. We received several letters, emails, and other documents that describe TFK's struggles. 21 TFK in various responses to EEO complaints, summarizes the problems they faced: Starting in the fourth quarter of 2010, and continuing thereafter, TFK began requesting [the] Union, Local 440, to provide miners to the Project site. TFK believed that the term miner was a term of art, generally recognized in the construction industry and by labor unions. Apparently, however, Local 440 did not share TFKs definition of the term miner. When Local 440 did not dispatch the caliber of miners TFK needed on the Project, the Joint Venture began identifying for Local 440 the characteristics of the miners TFK needed. Although the miners TFK needed to work on the Project in June, 2011 generally fell into the category of Group V/Group VI miners, Local 440 continued to dispatch an alarming number of unqualified Group V Miners in response to TFKs requests for qualified miners who possessed certain skills. TFK, therefore, was forced to implement an interview process to determine whether the individual(s) Local 440 dispatched as Group V/Group VI miners possessed the requisite skill set TFK needed its workers to possess to work on the Project. Once a dispatched miner arrived at the TFK Office, TFKs Site Superintendent, Bert Dore, conducted an interview with the prospective employee. 22
In the process of obtaining qualified workers, TFK took several extraordinary steps all of which underscore the serious problem they faced. They put out a call to the national 21 See footnote 11, Krulc, Michael A. Letter to Alan Clune, Laborers' Local 440, Seattle, WA. 19 Sep. 2011; Krulc, Michael A. Letter to Edwin F. Shorey, Seattle Tunnel and Rail Team (START), JV, Seattle, WA. 5 Jan. 2012.; Krulc, Michael A. Letter to Edwin F. Shorey, Seattle Tunnel and Rail Team (START), JV, Seattle, WA. 9 Jan. 2012.; Krulc, Michael A. Letter to Edwin F. Shorey, Seattle Tunnel and Rail Team (START), JV, Seattle, WA. 2 Feb. 2012.; Krulc, Michael A. Letter to Alan Clune, Laborers Local 440, Seattle, WA. 22 Mar. 2011.; Saperia, Jeremy. Sound Transit Project Seattle, WA. E-mail to Sergio Rascon. 6 Jan. 2012.; Saperia, Jeremy. FW: Helmets to Hardhats Termination. E-mail to Anne Braudis. 18 May. 2012. 22 TFK J/V response to request for information EEOC Charge No. #551-2011-01373 ECONorthwest Evaluation of Employment Outcomes at TFK J/V 15 union for qualified workers. 23 They implemented an interview process for dispatched workers which imposed significant time and opportunity costs on the project superintendent. 24 They have begun to self-certify shot-crete workers, bypassing the usual certification process. 25 All of these extra efforts have imposed additional costs on project management costs that economists would expect profit-maximizing firms to avoid unless absolutely necessary. This qualitative evidence combined with the quantitative evidence described previously clearly indicates that TFK had problems finding qualified laborers group 5 or higher. As described in section III, the disparities in outcomes between African Americans and other ethnic groups occurred within the exact categories of workers that TFK struggled to find qualified workers of any ethnicity (group 5/6 miners). This suggests that the racial disparity may simply be an extension of this larger problem; however, we need data describing individual qualifications to conclusively demonstrate this. C. Evidence from applications and EEOC responses While we do not have comprehensive data describing individual qualifications and performance, we do have some evidence that further suggests that such differences matter. First, TFK provided 838 pages of application and interview materials (including employment applications, resumes, a checklist where prospective workers indicated their experience, and notes made by an interviewer as they went through this check list) that suggest that differences in prior experience matter. These 838 pages included information for 78 (out of the 149) laborers in our data, but the information available for each individual differs. 26 For instance, some individuals have employment applications, skill checklist, interview notes, and a resume, but others only have a skill checklist or only an employment application. Without a consistent set of information for each person, we cannot use these records to perform a rigorous statistical analysis. However, systematic inspection of these documents reveals clear differences between the 9 laborers who were turned around, the 10 laborers who were dismissed due to inability to perform, and the remaining laborers for whom we have information. Specifically, individuals who provide clear evidence on their employment application or resume that they previously worked on a tunneling project are nearly always successful in obtaining employment and remaining on the job regardless of race. In contrast, none of the individuals turned around (for whom we have records) have unambiguous records indicating prior experience working on a tunneling project. Similarly, most of those dismissed due to inability to perform provide no evidence or cryptic evidence of prior tunneling experience.
23 Interview with Michael Krulc, April 25, 2012, Saperia, Jeremy. Sound Transit Project Seattle, WA. E- mail to Sergio Rascon. 6 Jan. 2012. 24 Interview of Bert Dore, September 27, 2011, Interview with Michael Krulc, April 25, 2012 25 Interview with Michael Krulc, April 25, 2012 26 The information also included a variety of individuals from other unions and individuals who do not appear to have been dispatched. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 16 To help make this description more concrete, Appendix A contains application materials for a few selected individuals. 27 Employee A, an African-American laborer, illustrates the type of applicant who is nearly always successful at obtaining a job and keeping it. Employee A provides a complete resume with detailed description of his prior experience, including work on two previous tunnel projects. He also checks nearly every box on the TFK certification and skill/experience checklist. Employee B, a Caucasian laborer, illustrates the type of applicant who is frequently turned around. The information we have for Employee B does not include an employment application or resume. We only have the TFK certification and skill/experience checklist, and, on this form, Employee B indicates that he possesses only two certifications and only 10 out of 23 skill/experience requirement. Employee C, a Hispanic laborer, provides another clear example of the type of worker who was turned around. Again, for Employee C we only have the certification and skills/experience checklist, and he possess three relevant certifications and only 3 out of 23 skill/experience requirement. Finally, Employee D, an African-American laborer, illustrates the type of worker dismissed due to inability to perform. Employee Ds information includes an application and two certification and skill/experience checklists. The employment application includes cryptic information on work experience, and his certification and skills/experience checklist indicates experience with 14 out of 23 categories. His records indicate he may have some experience, but the extent of his experience is unclear. The second type of qualitative materials that suggest that qualifications and experience matter are TFKs responses to EEOC complaints. These responses detail TFKs reasons for dismissing several employees, and these descriptions clearly indicate TFKs position that work performance was the reason for workers dismissal. We describe four examples obtained from these documents below. First, Employee E, an African- American Laborer, was dismissed due to inability to perform. Specifically, he was dismissed because he could not perform basic mining tasks and appeared unable to follow directions related to his own safety in the tunnel. 28 TFKs response further documents Employee Es failure to complete simple tasks correctly and his violation of a workplace order relating to no foot traffic in the tunnel. Second, TFKs response to a complaint by Employee F, an African-American laborer, documents complaints by Employee Fs co-workers regarding his work effort and competency that were corroborated by the observations of Employee Fs supervisor who noted two separate occasions where Employee F took inordinate amounts of time recovering his tool belt from different locations. 29 Third, TFKs response to the complaint of Employee G, an African-American laborer, documents that Employee G engaged in several unsafe practices the led a co-worker to describe him as an accident waiting to happen, that he failed to adequately complete assigned tasks (which led to co-workers having to re-do his work), and that he twice neglected to warn co-workers of approaching locomotive traffic. 30 Finally, TFKs response to Employee H documents that TFK gave Employee H
27 We redact specific names and personal identifying information for confidentiality purposes. 28 TFK response to Request for Information, EEOC Charge No. #551-2011-01391 29 TFK response to Request for Information, EEOC Charge No. #551-2011-00415 Amended 30 TFK response to Request for Response, EEOC Charge No. #846-2012-10533 ECONorthwest Evaluation of Employment Outcomes at TFK J/V 17 repeated warnings about not pulling his weight. He was even moved between crews in an effort to find a way to get him on the right track; however, ultimately, it was concluded that Employee H demonstrated not only a poor work ethic, which was undermining crew morale, but [he was] also a safety hazard. 31
Combined, the evidence from employment applications and EEOC responses clearly suggests that worker qualifications and performance matter. Workers whose records clearly demonstrate prior experience tend to be hired and remain employed, while those whose records do not clearly indicate prior experience are more likely to be turned around. Furthermore, TFK records clearly describe several cases where job performance, not race, led to workers being dismissed due to inability to perform after a relatively short time on the job. While we do not have the complete data on individual qualifications and performance required to statistically test the hypothesis that differences in qualifications and performance explain the observed differences in worker outcomes, the information we have reviewed is consistent with the view that qualifications and performance matter and that they explain any observed differences. D. Evidence from jobs employing same management in other locations While we do not have a complete data for the all of the individuals at issue in this matter, we do have additional data about the sponsoring contractor Traylor. As we understand, a substantial proportion of the supervisory team for the TFK job worked together for previous Traylor joint ventures. 32 This allows us to conduct an analysis that flips the logic of the Polissar and Neradilek UW vs CH analysis. Instead of comparing the outcomes for different contractors drawing workers from the same union, we examine the experience of the same contractor drawing workers from unions in a different cities. If we assume (a) the distribution of skills for the relevant workers differs across cities and (b) that a contractor who discriminates would do so consistently across jobs, then this analysis can help us distinguish among the competing stories for the observed differences in outcomes described above. If we do not observe differences between African American and other ethnic groups outcomes in other locations, than it is more plausible that the differences we observe in Seattle reflect differences in abilities and not discrimination. Specifically, we obtained data from two jobs completed in Los Angeles (the LA jobs) that employed the Traylor supervisory team currently working at the UW site the Northeast Interceptor Sewer Tunnel (the NEIS job) and the Metro Goldline Eastside Extension for LA Metro (the MTA job). 33 The data we were provided include ethnicity,
31 TFK response to Request for Information, EEOC Charge No. #551-2011-01373 32 Interview with Michael Krulc, April 25, 2012; email from Anne Braudis May 8, 2012 33 As we understand it several of the TFK management team also worked on these projects including TFK superintendent Bert Dore. Interview with Michael Krulc, April 25, 2012; email from Anne Braudis May 8, 2012 ECONorthwest Evaluation of Employment Outcomes at TFK J/V 18 job classification and title (including union membership), hire date, rehire date, final termination date, and total hours worked for each job. 34 Since the two jobs were largely (though not exactly) sequential, many people who worked on one job also worked on the other. For instance, nearly one third of laborers in the data worked on both jobs. In order to avoid difficulties related to people hired briefly on one job and subsequently transferred to another and difficulties related to uncertainty about the precise durations of employment, we combine the two datasets and examine total hours worked on both jobs as our primary outcome of interest. 35
Overall, worker outcomes on the LA jobs differ substantially from the outcomes observed in Seattle. First, zero workers were turned around at the LA jobs. 36 Second, relatively few workers appear to have been terminated due to their inability to perform. As noted previously, approximately 30 percent of laborers dispatched in Seattle were turned around or terminated for reasons related to individual performance, and the vast majority of these terminations occurred within 40 days of orientation. In total, approximately one-third of dispatched laborers in Seattle worked fewer than 200 hours through 2011. In stark contrast, in LA, only 11 percent of laborers worked fewer than 200 total hours. These facts further corroborate TFKs claim that they faced great and unusual difficulties obtaining qualified laborers in Seattle, and these facts provide support for one of the key assumptions underlying this analysis that the distribution of skills and qualifications among the dispatched workers from the relevant unions differ across cities. Examining differences across ethnic groups, we do not find evidence that African Americans experienced worse outcomes than other ethnic groups at the LA jobs. Table 7 presents average total hours 37 , percentage working fewer than 200 hours, and share of NEIS workers retained for the MTA job by ethnic group for laborers. Relative to Caucasians, African Americans work more total hours, were less likely to work fewer than 200 hours, and were more likely to move from the NEIS job to the MTA job. None of these differences though are statistically significant at the commonly used 0.05 level. In fact, the differences in average hours and share working fewer than 200 hours have very large confidence intervals (or very high p-values). This stems from both the small differences in values and the relatively small numbers of African Americans in the data. Only 10 African American laborers were dispatched for these jobs. While the small sample of African Americans limits our ability to draw strong conclusions from these data, the patterns we do observe do not suggest that the common Traylor supervisory team discriminated against African Americans over the past decade.
34 The data for these analyses come from Traylor Brothers payroll system and were forwarded to ECONorthwest on May 8, 2012. 35 We also do not have the date for all terminations we observe when people are rehired, but we do not observe their original termination date (only their final termination date). 36 This finding from the data was confirmed by Jeremy Saperia. 37 We limit the sample in the total hours analysis to workers hired on date equal to or later than the first African American. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 19 Table 7. Summary statistics for LA jobs by Ethnicity Average Total Hours Percent working <200hours Percent of NEIS workers retained for MTA Caucasian 1559 18% 26% African American 1650 10% 60% P-value for difference 0.89 0.57 0.07 Source: ECONorthwest analysis of data provided by Traylor Brothers Figure 1 helps to illustrate a key theme of our analysis the set of workers where we have observed statistical disparities in worker outcomes across ethnicity represent a small share of the total set of workers employed by the TFK management team. Each rectangle in the figure is proportional to the share of all workers across the LA and Seattle jobs in each of the labeled categories. As the figure indicates Group 5 and 6+ laborers for the Sound Transit job (the grey areas) represent a small share of the total workers across these jobs (less than 20%). Outside the upper left-hand corner (i.e., across a majority of the relevant workers), we do not observe large or statistically significant differences in outcomes. Figure 1. Visualization of the share of total workers on Seattle and LA jobs for whom we observe disparities in worker outcomes by ethnicity
All WesL CoasL
Sound 1ranslL (S1) ro[ecL
CLher S1 Laborers S1 Laborers, Croup S S1 Laborers, Croups 6+ ECONorthwest Evaluation of Employment Outcomes at TFK J/V 20 V. CONCLUSION In sum, the data available do not support a conclusion that TFK discriminates against minorities, particularly African Americans. The data produced by Sound Transit contain numerous inconsistencies which may affect the accuracy and reliability of any data analysis, particularly an analysis that includes relatively few data points. Setting aside the data issues, data analysis reveals that statistically significant discrepancies in worker outcomes are limited to one worker classification from a single union Group 6 tunnel miners from the laborers union. 38 Statistical analysis also suggests that at previous jobs in a different city that employed an overlapping supervisory team, African Americans did not experience worse outcomes than Caucasians. Thus, to assume that this evidence indicates discrimination by TFK one must believe: (1) The underlying distribution of qualifications and skills was the same across ethnic groups for individuals dispatched to the TFK site (even though we are aware of no evidence that support this assumption and the limited available evidence contradicts it) (2) That the source of the discrimination only works with or discriminates against African Americans who are laborers in group 6 (or possibly 5) and higher (3) That the source of discrimination is unique to the Seattle project (or was not present during the LA projects). While we cannot rule out the possibility that all three of these beliefs may be true, this appears implausible. Given that the affected subset of a subset of workers was precisely the group with which TFK experienced significant difficulties finding qualified individuals, it seems more plausible that the observed statistical discrepancy reflects unobserved differences in individual ability, rather than racial discrimination.
38 Differences between African Americans and Caucasians in group 5 may be significant if one expands the threshold to determine statistical significance beyond the commonly used 0.05 level. ECONorthwest Evaluation of Employment Outcomes at TFK J/V 21 APPENDIX A SELECTED APPLICATION AND INTERVIEW MATERIALS ECONorthwest Evaluation of Employment Outcomes at TFK J/V 22 Employee A ECONorthwest Evaluation of Employment Outcomes at TFK J/V 23 Employee A ECONorthwest Evaluation of Employment Outcomes at TFK J/V 24 Employee A ECONorthwest Evaluation of Employment Outcomes at TFK J/V 25 Redacted Employee A ECONorthwest Evaluation of Employment Outcomes at TFK J/V 26 Employee B Redacted ECONorthwest Evaluation of Employment Outcomes at TFK J/V 27 Employee C ECONorthwest Evaluation of Employment Outcomes at TFK J/V 28 Employee D ECONorthwest Evaluation of Employment Outcomes at TFK J/V 29 Employee D