Inequality at Work: The Effect of Peer Salaries on Job Satisfaction

Transcription

1 Inequality at Work: The Effect of Peer Salaries on Job Satisfaction David Card, UC Berkeley Alex Mas, Princeton Enrico Moretti, UC Berkeley Emmanuel Saez, UC Berkeley April

2 MOTIVATION Possibility that individuals care about relative income is an old idea in economics (Adam Smith 1759, Veblen 1899, Duesenberry, 1949) Relative income concerns are critical for relevance of inequality and for policy Large body of empirical evidence finds correlations of relative income and (a) job satisfaction (Clark and Oswald 96) [also affects wage setting, Bewley 99], (b) happiness (Luttmer 05), (c) health (Marmot 04), (d) reward-related brain activity (Fliessbach et al. 07) Many lab experimental studies also find evidence of relative income effects 2

3 INTRODUCTION Credible field evidence is hard to obtain as peer group incomes are highly endogenous This paper: Instead of manipulating peers salary, we manipulate information on peers salary (a) We informed a random subset of UCSC, UCSD, UCLA workers about a new website listing pay of all UC workers; (b) We surveyed both control and treatment employees Key results: 1. Asymmetric response: (a) treated employees paid below median in their unit report being less satisfied with job and more likely to search for new job, (b) No effect for treated employees above median 2. No effect on medium-term turnover (so far) 3

4 OUTLINE 1. Two alternative theoretical models: Relative income concerns Learning about future pay from co-worker pay 2. Experimental design, data, selection issues 3. Empirical Results First Stage effects on website use Effects on job satisfaction and job search intentions Actual turnover after 2-3 years 4

5 MODEL 1: RELATIVE INCOME CONCERNS Satisfaction S S(w, I) = u(w) + v(w E[m I]) + e depends on own pay w and perceived gap between w and median=mean pay m in peer group based on information I. u(.) and v(.) increasing and normalized u(0) = v(0) = 0 Assuming first complete initial ignorance and complete knowledge when treated Control group: E[m I 0 ] = w Treatment group: E[m I 1 ] = m S(w, m, D) = u(w) + D v(w m) + e where D = indicator for informed status 5

6 MODEL 1: SPECIFICATIONS (a) Linear relative income concerns: v(w m) = b (w m) (b) Inequality aversion (Fehr-Schmidt QJE 99): v(w m) = b 0 (w m) 1(w m) + b 1 (w m) 1(w > m) with b 0 > b 1 0 (c) Rank based income concerns (Parducci 95): v(w m) = b 0 [rank(w).5] 1(w m)+b 1 [rank(w).5] 1(w > m) with b 0 > b 1 0 Implications: (1) Treatment lowers S if w < m, (2) Treatment (weakly) increases S if w > m, (3) Average treatment effect is zero in (a) or weakly negative in (b-c) 6

7 MODEL 2: LEARNING FROM PEER PAY At UC, growth in pay is strongly related to relative pay in ones pay unit defined as department (faculty vs. staff) (a) Employees paid below median experience higher pay growth both in percentage and in absolute dollars (b) Basic OLS regression on year-to-year pay growth: w = α + β w + γ m + ε generates γ > 0 (and β < 0) both in level and log specifications Being paid below median is a positive signal on future pay growth 7

8 A1. Earnings Growth and Median Peer Earnings Below/Above Median comparison Regressions Own 2007 pay Median 2007 pay Below Median Above Median Above - Median A. Log pay changes from 2007 to 2008 log (total pay) (.007) (.006) (.009) log (base pay) (.007) (.007) (.011) B. Level pay changes from 2007 to 2008 (total pay) $6,775 $4,906 -$1, (433) (.004) (.006) (base pay) $6,246 $4,142 -$2, (262) (.004) (.006) Observations ,159 13,159 13,159 Notes: Sample is all UCLA employees matched to their earnings and who are present in both 2007 (base year) and 2008 (following year). Pay unit (for median computation) refers to faculty or staff members in an individual's department.

9 MODEL 2: LEARNING FROM PEER PAY S(w, I) = w + βe[w I] + e where w is current pay and w future pay Prior w N(w, 1/q) E[w I 0 ] = w Mean pay of co-workers m is a noisy signal of future pay: m = w + u, u N(w, 1/k) Posterior after seeing signal is: E[w I 1 ] = (1 λ)w + λm with λ = k/(q + k) S(w, m, D) = (1 + β)w + D (βλ) (m w) + e Same as linear relative income concerns but with opposite sign 9

10 INCOMPLETE COMPLIANCE We have: S = u(w) + D v(w m) + e Treatment T affects probability of being informed D: π 0 = E[D T = 0, w, m] 0.25 π 1 = E[D T = 1, w, m] 0.5 S = u(w) + π 0 v(w m) + T (π 1 π 0 ) v(w m) + e + φ with E[φ w, m] = 0 If π 1 π 0 is constant across people we have simple attenuation Critical to verify that the effect of T on use of the website does not vary with w, m to identify v(w m). 10

11 EXPERIMENT DESIGN California State workers pay was public but hard to access [required written letter] Following right-to-know court decision, Sacramento Bee newspaper obtained complete micro-data and posted it online in March 2008 as a searchable database Database initially included name, position, agency, 2007 annual earnings (base and total) for all UC workers [first time this was done] Randomized information experiment done in October 2008 for UCSC, November 2008 for UCSD, May 2009 for UCLA Data: (1) 2008 online directories, (2) Admin earnings data, (3) follow-up survey, (4) 2011 online directorories 11

12

13

14 EXPERIMENT DESIGN First Stage: Information Treatment Collect directory data online in September Alert a random subset of state employees about the SacBee website [stratified by department at UCSC, UCSD, UCLA] Second Stage: Online Survey Invite (by ) state employees to fill out a short survey (12 questions) on job satisfaction, perceptions on pay equity, and job search plans [with randomized prize incentives] 3-10 days after 1st stage Third Stage: Medium-run turnover effects Collect directory data in March 2011 to assess turnover effects 13

15 TREATMENT Subject: Pay Inequality at UC We are Professors of Economics at Princeton University and Cal Berkeley conducting a research project on pay inequality at the University of California. The Sacramento Bee newspaper has launched a web site listing the salaries for all State of California employees, including UC employees. The website is located at or can be found by searching Sacramento Bee salary database with google. As part of our research project, we wanted to ask you: Did you know about of the Sacramento Bee salary database website? 14

16 PLACEBO At UCLA (only), we also sent a Placebo to a subgroup within the control The placebo has exactly the same structure as the treatment but informed people about a UC website listing the salaries of top UC administrators instead of the SacBee website Placebo carried out to test that receiving an about Pay Inequality does not create priming effects independently of learning about peers salaries 15

17 SECOND STAGE SURVEY Survey sent to all employees with days after treatment Prize for completing the survey offered to a random subset of employees ($1000 for 3 respondents in each campus) [exogenously shifts response rate, used to check selection issues] Survey completed online by following link [we can match respondents to their and admin compensation data] Survey includes questions on SacBee website use, job satisfaction, pay satisfaction, pay fairness at UC, job search intentions, views on overall inequality, basic demographics 16

18 1. Design of the Information Experiment Campus Information Treatment Response Incentive ($1000 prize) UC Santa Cruz 2/3 depts treated 1/3 depts -- all offered N=3,606 60% of workers in treated 1/3 depts -- 1/2 offered 223 departments depts get treatment 1/3 depts -- none offered target = 40% of workers target = 50% of workers actual = 42.0% actual = 49.3% UC San Diego 1/2 depts treated 1/3 depts -- all offered N=17,857 50% of workers in treated 1/3 depts -- 1/2 offered 410 departments depts get treatment 1/3 depts -- none offered target = 25% of workers target = 50% of workers actual = 23.9% actual = 55.0% UCLA 1/2 depts treated All offered N=20,512 75% of workers in treated 445 departments depts get treatment target = 37.5% of workers actual = 36.4% All Three campuses target = 32.4% of workers target = 74.4% of workers N=41,975 actual = 31.6% actual = 76.5% 1078 depts

19 A2: Matching and Response Rates % No. in % Matched Responded In final sample (wage + survey) Directory to Pay to Survey Percent # obs. (1) (2) (3) (5) (6) Staff 34, ,419 Faculty 7, ,018 All 41, ,437

20 2. Determinants of Survey Response All Coefficients in percent Full Sample (N=41,975) Matched to Wage Data (N=31,887) Dummy if match to wage (0.58) Treatment Effects: Treated individual (0.70) (0.79) (0.78) Untreated ind. in treated dept (0.82) (0.92) (0.91) -- Placebo Treatment Effects: Placebo individual (1.05) (1.20) (1.17) (1.01) Response Incentive Effects: Offered prize in 100% dept (0.99) (1.11) (1.10) (0.86) Offered prize in 50% dept (1.18) (1.38) (1.36) -- Not offered prize in 50% dept (1.29) (1.43) (1.46) -- Treatment Effects Based on Relative Wage: Treated individual with w m in pay unit (0.79) Treated individual with w > m in pay unit (0.81) Dummy if w > m in pay unit (0.75) Cubic in wage? no no yes yes

21 3. Comparison of Treated and Non-treated Individuals Mean of Mean of Difference Control Treatment (adj. for t-test Group Group campus) Overall Sample (N=41,975) Percent faculty Percent matched to wage data Sample Matched to Wage Data (N=31,887) Mean base earnings ($1000's) Mean total earnings ($1000's) Percent with total earnings < $20, Percent with total earnings > $100, Percent responded to survey Survey Respondents with Wage Data and non-missing Values (N=6,411) Percent faculty Mean total earnings ($1000's) Percent female Percent age 35 or older Percent employed at UC 6 years or more Percent in current position 6 years or more

22 4a: Percent Effect of Treatment on Use of Sacramento Bee Website Dependent variable is SacBee website use Treated individual (1.8) (1.6) (1.6) (1.6) (2.1) Untreated ind. in treated dept (1.7) Treated ind. with w m (2.2) Treated ind. with w > m (2.0) Treated individual (w-m) (0.7) Treated individual (w-m) 1(w-m>0) (1.0) Dummy for response incentive (test for selection bias in respondent sample) (1.8) Dummy for w m (1.9) Deviation of wage from median (0.50) Deviation of wage from median if deviation positive (0.60) Demographic controls no no no yes yes yes P-value for test against model in col Mean of SacBee use for control group is 24%

23 All results in % 4b. Which Salaries People Check on SacBee Website? Overall Use SacBee website Workers in own dept Workers in other depts, own campus Workers at other campuses "Highprofile" UC workers Any of those in cols. 2-5 Mean rate of use for control group Estimated treatment effect from model with basic controls: Treated individual (2.4) (2.2) (1.7) (1.4) (2.0) (2.4) Estimated treatment effect from interacted model with basic controls: Treated individual with w m (3.5) (3.3) (2.3) (2.0) (2.9) (3.5) Treated individual with w > m (2.8) (2.7) (2.1) (1.7) (2.4) (2.8) P-value for equality of treat. effects

24 FIRST STAGE FINDINGS 1. Informing workers about the website has a large impact on information on peers salary: Treatment doubles the rate of use from 25% to 50% 2. Effect on rate of use is uniform across pay groups Assumption that π 1 π 0 constant seems to hold so we can identify effects by pay level 3. Most new users (80%) report that they investigated the wages of colleagues in their own department Dept seems relevant peer unit We will define pay unit = Department (faculty/staff) 4. No spillover of treatment within departments 19

25 SECOND STAGE SURVEY QUESTIONS 1. How satisfied are you with your wage/salary on this job? [0-3] 2. How satisfied are you with your job? [0-3] 3. Do you agree or disagree that your wage is set fairly in relation to others in your department/unit? [0-3] 4. How likely is it you will make a genuine effort to find a new job within the next year? [0-2] 5. Are differences in income in American too large? [0-3] 1-3 combined into an overall 10 point scale Satisfaction Index [for simplicity and precision] 1-4 combined into a 0-1 Dissatisfied and likely looking for new job 20

26 A3. Means of Outcome Measures by Treatment Status "How satisfied are you with your wage/salary on this job?" "How satisfied are you with your job?" "How likely is it you will make a genuine effort to find a new job within the next year?" "Do you agree that your wage is set fairly in relation to others in your dept/unit?" "Do you agree or disagree that differences in income in America are too large?" Not At All Satisfied Not Too Satisfied Somewhat Satisfied Very Satisfied All (N=6411) Control Reweighted Treatment Group All (N=6411) Control Reweighted Treatment Group Not At All Likely Somewhat Likely Very Likely All (N=6411) Control Reweighted Treatment Group Strongly Disagree Disagree Agree Strongly Agree All (N=6411) Control Reweighted Treatment Group All (N=6397) Control Reweighted Treatment Group

27 SECOND STAGE ORDERED PROBITS SPECIFICATIONS S = g(w, x) + b T + e S = g(w, x)+a 1(w m)+b 0 T 1(w m)+b 1 T 1(w > m)+e S = g(w, x)+b 0 T (w m) 1(w m)+b 1 T (w m) 1(w > m)+e S = g+b 0 T [rank(w).5] 1(w m)+b 1 T [rank(w).5] 1(w > m)+e We always include controls for campus (faculty/staff) and cubic in w We sometimes add demographic controls x (gender, age, tenure) We always cluster s.e. by pay unit = dept (faculty/staff) 22

28 5a. Effect of Information Treatment on Job Satisfaction/Search Dissatisfied and Satisfaction Index Likely to Look for New Job Likely Looking for a New Job (10 pt scale) (1-3 scale) (0-1) Treated individual (3.3) (3.4) (5.1) I. Treated ind. with w m (4.5) (4.5) (4.5) (4.5) (6.6) (6.6) II. Treated ind. with w > m (4.1) (4.1) (4.9) (4.6) (7.5) (7.5) II-I (5.4) (5.4) (6.6) (6.4) (9.6) (9.5) Demographic controls No No Yes No No Yes No No Yes P-value for exclusion of treatment effects

29 A4: Treatment Effect on Measures of Job Satisfaction Wage is fair Satisfied with Wage on Job Satisfied with Job Likely to Look for New Job (1-4 scale) (1-4 scale) (1-4 scale) (1-3 scale) I. Treated ind. with w m (4.9) (4.5) (4.9) (4.5) II. Treated ind. with w > m (4.5) (4.5) (4.4) (4.9) II-I (6.0) (5.7) (6.5) (6.6) P-value for exclusion of treatment effects

30 6a. Treatment Effects: Income Gap vs. Rank Satisfaction Index Likely to Look for New Job Dissatisfied and Likely Looking for a New Job (10 point scale) (1-3 scale) (0-1) Treated ind. (w-m) 1(w m) (1.4) (2.4) (1.5) (2.5) (2.2) (3.6) Treated ind. (w-m) 1(w > m) (1.0) (1.6) (1.1) (1.7) (1.9) (2.4) Treated ind. (rank-.5) 1(w m) (1.5) (2.8) (1.7) (2.9) (2.3) (4.0) Treated ind. (rank-.5) 1(w>m) (1.4) (2.5) (1.7) (2.7) (2.9) (4.2) P-value for exclusion of treatment effects w= own pay. m = median pay in pay unit

31 SECOND STAGE FINDINGS 1. Information has a slightly negative effect on overall utility. 2. Average masks strong heterogenous effects: (a) Treatment reduces job satisfaction and increases job search intentions for workers below median, (b) no significant change for workers above median Effects (a) are large: being closer to median by $10K in Treatment $5K extra pay in control 3. Rank relative to median seems to matter more than pay gap relative to median Results support relative income model based on rank (Parducci 95) and with nonlinearity (Fehr-Schmidt 99) and go against learning about future pay model 24

32 SELECTION ISSUES Concern that survey respondents are different across treatment and control groups: e.g., treatment induces disgruntled employees to answer We have verified that treatment and control respondents are identical along observables (wage, gender, age, tenure, etc.) Randomized prize incentives and placebo shifted response rate and should have no direct impact on job satisfaction/search intentions: can be used in a two-step Heckit procedure We find no correlation between response dummy and outcomes but results depend on functional forms and not very stable 25

33 5b. Test for Selection Issues using Prize and Placebo Exclusions Satisfaction Index (10 pt scale) Heckit Likely to Look for New Job (1-3 scale) ML Select. Model Dissatisfied and Looking for a New job (0-1) ML Select. Model I. Treated ind. with w m (4.5) (2.9) (4.5) (5.0) (6.6) (7.8) II. Treated ind. with w > m (4.1) (2.8) (4.9) (5.3) (7.5) (8.0) II-I (5.4) (3.6) (6.6) (6.5) (9.6) (9.6) Inverse Mills Ratio (0.14) Correlation (0.25) (0.32) P-value for exclusion of treatment effects Heckit is a two-step Heckman selection model. ML Select. Model are maximum likelihood selection models for an ordered categorical dependent variable. Placebo + Prize affect response rate but excluded from Second Stage.

34 EFFECTS BY SUBGROUPS We can run basic specs for specific subgroups (faculty vs. staff, men vs. women, high vs. low tenure) 1. Female, staff, and low tenure highly responsive along job search intention (men, faculty, high tenure are not) 2. High tenure and staff particularly responsive along satisfaction index (faculty and low tenure less responsive) 3. Faculty are highly responsive when median is defined at campus level both below and above median (humanities resent econ/business/law, econ/business/law feel better after seeing humanities) 27

35 A6. Information Treatment Effects -- by Subgroup A. Satisfaction Index (10 point scale) Female Male Staff Faculty Low Tenure High Tenure I. Treated ind. with w m (5.5) (6.8) (5.3) (9.9) (6.1) (6.3) II. Treated ind. with w > m (5.6) (6.2) (4.5) (9.6) (7.5) (4.8) II-I (7.5) (8.6) (6.3) (13.9) (9.0) (7.4) P-value for exclusion of treatment effects Observations

36 A6. Information Treatment Effects -- by Subgroup B. Likely to Look for New Job (1-3 scale) Female Male Staff Faculty Low Tenure High Tenure I. Treated ind. with w m (5.7) (8.2) (5.2) (10.8) (6.4) (6.9) II. Treated ind. with w > m (6.4) (7.0) (5.5) (11.7) (8.5) (5.6) II-I (8.1) (11.4) (7.3) (15.2) (10.3) (9.1) P-value for exclusion of treatment effects Observations

37 A6. Information Treatment Effects -- by Subgroup C. Dissatisfied and Looking for a New Job (0-1) Female Male Staff Faculty Low Tenure High Tenure I. Treated ind. with w m (7.8) (11.1) (7.3) (15.4) (8.4) (9.6) II. Treated ind. with w > m (9.9) (10.2) (8.2) (18.9) (11.4) (9.1) II-I (12.1) (15.8) (10.3) (23.8) (13.5) (13.2) P-value for exclusion of treatment effects Observations

38 7.Treatment Effect by Pay Relative to Campus/Occupation Median Satisfaction Likely to Look Index for New Job (10 point scale) (1-3 scale) Dissatisfied and Likely Looking for a New Job (0-1) Faculty Staff Faculty Staff Faculty Staff I. Treated ind. with w m with m CAMPUS median (10.4) (5.3) (11.1) (5.3) (15.6) (6.8) II. Treated ind. with w > m with m CAMPUS median (8.9) (4.4) (11.2) (5.5) (19.6) (8.8) II-I (14.0) (6.0) (15.0) (7.4) (24.8) (10.5) P-value for exclusion of treatment effects w= own pay. m = median pay in pay unit defined as campus x (faculty/staff)

39 ADDITIONAL RESULTS 1. Placebo has no effect on job satisfaction / job search outcomes (a) Priming effects [ about Pay Inequality ] cannot explain the main treatment results (b) reinforces our confidence that main results are not driven by selection [as both treatment and placebo reduced survey response rate] 2. Treatment increases concerns about overall inequality in America both below and above median 29

40 8. Estimates of the Effect of "Placebo" Treatment Treat ment Placebo p- value Likely to Look for New Job Treat ment Placebo p- value Dissatisfied and Likely Looking Satisfaction Index (10 point scale) (1-3 scale) for a New Job (0-1) Treat ment Placebo p- value I. Treated ind. with w m (7.2) (7.2) (7.4) (9.6) (9.5) (16.4) II. Treated ind. with w > m (6.1) (6.1) (7.3) (7.5) (10.7) (11.0) Observations 2,303 1,880 2,303 1,880 2,303 1,880 "Treatment" in the columns denotes the information treatment. "Placebo" denotes the placebo treatment. P-value tests difference between treatment and placebo effects. Sample is UCLA only.

41 9. Treatment Effect on Perceptions of Overall Inequality Differences in Income in America Are Too Large (1-4 Scale) (1) (2) (3) (4) Treated individual (3.7) (3.7) (4.7) (4.7) Treated ind. with w m -1.5 (6.2) -1.4 (6.2) Demographic controls No Yes No Yes

42 THIRD STAGE: EFFECTS ON ACTUAL TURNOVER In March 2011, we collected again directory information to define a medium-run turnover dummy. Results: 1. Turnover very correlated to job search intention self-report 2. No effect of treatment on medium-run turnover (re-assuring to find no effect in responder sample as survey asked about SacBee) Possible reasons for no effect on turnover: 1. SacBee (and others since then) information has diffused in 2-3 years to both control and treatment 2. Great recession sharply reduced job mobility opportunities 32

43 10. Treatment Effect on Actual Job Mobility Non Responders responders Full sample (N=6,835) (N=25,048) (N=31,883) Reported "very likely" to make 20.5 effort to find a new job (1.6) Reported "somewhat likely" to 5.9 make effort to find a new job (1.1) Treated individual (1.3) (1.2) (1.1) Treated ind. with w m (1.9) (1.5) (1.4) Treated ind. with w > m (1.5) (1.2) (1.1) Dependant variable is 1 if able to match to online campus directories in

44 CONCLUSIONS 1. Our treatment design was effective in providing information about peers pay 2. Asymmetric effects: (a) treated employees paid below median in their unit report being less satisfied with job and more likely to search for new job, (b) No effect for treated employees above median 3. No effect on medium-term turnover (we will explore effects on pay increases next) Relative pay matters with an asymmetry below vs. above median (as in inequality aversion models) 34

45 IMPLICATIONS AND FUTURE WORK 1. Employers have a strong incentive to impose pay secrecy rules Right-wing view: Forcing pay disclosure would reduce pay satisfaction in aggregate keeping salaries constant Left-wing view: But pay disclosure may induce workers to organize and ask for a flatter pay structure (Frank 84, Bewley 99) 2. Future work: Treat people with a left-wing vs. right-wing randomized information on US inequality and tax structure and assess whether it affects preferences for redistribution or voting behavior 35