The Economic Policy Institute Minimum Wage Simulation Model uses microdata from the American Community Survey (ACS), compiled by Ruggles et al. (2018), and from the Current Population Survey Outgoing Rotation Group (CPS-ORG) from the U.S. Census Bureau to estimate the impact of proposed changes in federal, state, and local minimum wages. This document describes how the model works, the data sources it employs, and the assumptions that are built into the model.
The model begins by creating a sample of wage earners from the 2013–2017 ACS that is demographically representative of the employed population at the state or congressional district level. First, the sample is restricted to those ACS respondents ages 16 and older who reported being employed in the week prior to being surveyed and who reported having positive wage income during the previous 12 months. Workers in the armed forces are also excluded.
Workers are first assigned state geography based on PUMA (Public Use Microdata Area) of residence. For national- and state-level analyses, the model uses PUMA of residence to assign the state of residence. For congressional-district-level analyses, because PUMAs can correspond to multiple congressional districts, the model assigns respondents one or many congressional districts but proportionally adjusts the ACS sample weights according to 2010 Census population allocation factors (MCDC 2018).
Because the employed population who also had earnings last year is more attached to the labor force than the typical employed worker, the model further recalibrates the sample weights to match estimated counts of certain demographic characteristics published at the state or congressional district level in the five-year 2013–2017 ACS tables. In particular, the model adjusts sample weights by iterative proportional fitting, or raking, to match state-level or congressional-district-level employment counts of those ages 16–19, 20–24, 25–34, 35–44, 45–54, 55–64, and 65 and older; those who are white, black, or nonwhite/nonblack, ages 16–64 and 65 and older; those who are or are not Hispanic, ages 16–64 and 65 and older; and those with education less than a high school degree, a high school degree, some college education, or an advanced degree (across all ages, 16 and older). This weight calibration is especially important in the case of congressional districts, where the naive PUMA-congressional district match can be inaccurate and where district boundaries can be determined in part by certain demographic characteristics.
This calibration procedure creates two sets of weights: one for state-level estimates (which we also use for national-level analyses) and one for congressional-district-level estimates. Because the weights differ, aggregating congressional-district values up to the state or national level may not match estimates produced using the state-level weights. However, these differences are small. For example, aggregating the values for all affected workers within each demographic group across all congressional districts in EPI 2019 produces totals that differ by no more than 0.3 percent from the national (state-weighted) estimates in Cooper 2019.
After recalibrating weights, the model imputes hourly wages in three steps: using previous year earnings, using predicted hourly wages based on respondents’ characteristics and CPS-ORG data, and, finally, using CPS-ORG state-level wage distributions.
After excluding from the above sample any self-employed workers and any individuals whose place of work is abroad, the model first calculates an initial hourly wage based on earnings over the prior 12 months, divided by the product of usual hours worked per week and weeks worked in the last 12 months. Because the ACS provides only binned information on weeks worked, the model imputes weeks worked within each bin using an out-of-sample linear prediction from March Current Population Survey data from the same sample years. The weeks-worked bin-specific OLS prediction of weeks worked last year is conditional on dummy variables for gender, education, race and ethnicity, marital status, major industry, major occupation, part-time status, a quintic polynomial in age, and year and state fixed effects.
Second, the model then takes an average of the initial hourly wage above and the hourly wage predicted by a linear regression from respondents in the CPS-ORG (similar to the hourly wage construction in CBO 2014). The model regresses the CPR-ORG hourly wage (reported or calculated as weekly earnings divided by weekly hours) on dummy variables for gender, education, race and ethnicity, marital status, major industry, major occupation, part-time status, a quintic polynomial in age, and year and state fixed effects. Then the model uses that fitted equation to predict hourly wages in the ACS sample. The mean of the initial and predicted hourly wage is the second-stage hourly wage.
Third, because mismeasurement of wages in the ACS causes the wage distribution to have abnormally long tails, the model creates the final, third-stage hourly wage based on the CPS-ORG state-level wage distribution for 2017. Specifically, the model determines the state- and year-specific percentile location of the ACS second-stage hourly wage, and then assigns the CPS-ORG hourly wage percentile value of that state- and year-specific wage quantile. For example, if a respondent in the 2014 ACS resides in California and has a second-stage hourly wage that is at the 12th percentile in the 2014 ACS California sample, then that respondent is assigned the 12th-percentile hourly wage of the 2017 CPS-ORG California sample. This third-stage hourly wage value completes the model’s wage imputations.
EPI’s simulation model incorporates a compiled data set of all applicable minimum wage and tipped minimum wage rates for all states, by month and year, from January 1984 onward. The model also incorporates data on all local (city, county, regional) minimum wages and tipped minimum wages from January 2009 onward. Minimum wage rates for states and localities with scheduled increases are incorporated and those with automatic annual indexing for inflation are projected using CBO projections for inflation as measured by the Consumer Price Index for all Urban Consumers (CPI-U), published in the Congressional Budget Office’s annual Budget and Economic Outlook report (see CBO 2018).
Indexing provisions in some state and local minimum wage laws prevent any increase in the minimum wage under certain labor market or state fiscal conditions, such as during periods of elevated unemployment or when state revenues decline over a given period. The EPI model assumes all scheduled or indexed increases will take effect as scheduled. Also, some minimum wage laws set different increase schedules for businesses of different sizes. In all cases, the EPI model assumes that the highest minimum wage possible (typically the minimum wage that would apply to the largest firms) will apply to all workers in the applicable jurisdiction. This assumption makes our estimates more conservative, as it is possible that a proposed minimum wage law would affect some workers whom our model assumes are already subject to a higher minimum wage than they actually are.
At the outset of any simulation, the model first adjusts wage values to reflect “natural” nominal wage growth—i.e., wage growth that would likely occur due to market forces. To do so, the model increases wage values by the average rate of nominal annual wage growth of the bottom 20 percent of wage earners in each state over the most recent five-year period using data from the CPS-ORG, prorated for the number of months occurring between the midpoint of the data period and the month of the proposed minimum wage increase. For example, the estimates in Cooper 2019 use 2017 ACS data to simulate a minimum wage increase first occurring in July 2019. Thus, the model adjusts these 2017 wage values by the state-specific five-year average from 2012 to 2017 of the annual nominal wage growth of the bottom 20 percent of wage earners, as calculated from the CPS-ORG, prorated for the 24 months between July 2017 (the midpoint of the data year) and July 2019 (the month of the first simulated minimum wage change). Because the wage growth reported in the CPS-ORG is based on workers’ place of residence, natural wage growth adjustments in the model are applied based on workers’ reported state of residence.
The EPI model then accounts for the impact of any scheduled minimum wage changes between the data period and the first simulated minimum wage increase by adjusting any wage values less than or equal to 115 percent of the existing minimum wage in areas where the minimum wage changes. For example, when simulating the impact of a minimum wage increase that would occur in July 2019 using 2017 data, all individuals in localities that raised their minimum wages between December 2017 and July 2019 would already have higher wages than recorded in the source data. In these localities, wage values below the scheduled or expected (in the case of an automatic inflation adjustment) minimum wage are increased in direct proportion to the expected minimum wage for all observations with wages equal to 115 percent or less of the existing minimum wage. For example, when simulating a proposed federal minimum wage increase that would occur in July 2019, if someone working in Colorado in August 2017 was earning $10 per hour, their wage was equal to 107.5 percent of the August 2017 state minimum wage ($9.30). That person’s wage would be adjusted to $11.94 per hour—107.5 percent of the expected Colorado minimum wage for July 2019 ($11.10).
It’s important to note that the ACS is one of few publicly available data sets that capture information on respondents’ geography of work in addition to their geography of residence. For modeling the impact of proposed minimum wage changes, this information is critical, as many people do not necessarily work and live in the same city, county, or even state. The minimum wage that applies in their geography of residence may be different from the one that applies where they work. The EPI Minimum Wage Simulation Model takes these differences into account, by making wage adjustments and identifying workers affected by the proposed minimum wage change based on respondents’ geography of work. In Cooper 2019, the state values reported reflect the impact on workers in those states, including those workers who live in other states. In other words, the reported state data are all based on the place-of-work PUMA of respondents. EPI (2019) shows impacts of the Raise the Wage Act of 2019 by congressional district. Because congressional representation is based on residence, the values reported show impacts based on respondents’ congressional district of residence, although the model still accounts for applicable minimum wages based on place of work.
The model also assumes population growth between the data period and the proposed first increase. For national-level simulations, person weights are adjusted by the projected annual labor force growth rate from 2014 to 2024 for specific racial groups as reported in BLS 2015. According to the Bureau of Labor Statistics projections, from 2014 to 2024, the white, non-Hispanic labor force will decline by 0.3 percent annually; the black, non-Hispanic labor force will grow by 1.0 percent annually; the Hispanic labor force will grow by 2.5 percent annually; the Asian labor force will grow by 2.1 percent annually; and the labor force of all other races will grow by 2.0 percent annually. These annual growth rates are adjusted by the number of months that occur between the midpoint of the data and the month that the first proposed minimum wage increase would occur. For state- or local-level simulations, growth rates are taken from whatever population or labor force projections exist for the applicable geography, typically from state labor or economic development agencies or from university-based demographic centers.
Having made these adjustments, the model then identifies “directly affected” workers as those workers in localities where the prevailing minimum wage (i.e., the higher of the local, state, or federal minimum wage) is less than the proposed minimum wage and whose hourly wage is greater than or equal to 80 percent of the prevailing minimum wage but less than the proposed minimum wage. This lower bound is set for affected workers at 80 percent of the existing minimum wage to allow for some measurement error in the hourly wage data, but also to exclude those workers who are either unlikely or ineligible to be affected by the change in the minimum wage. These could be workers who are exempt from minimum wage laws or who are victims of wage theft. In either case, if workers are reporting hourly wages less than 80 percent of the existing minimum wage, the model assumes that whatever is preventing them from being paid the existing minimum wage would likely prevent them from benefiting if the minimum wage were raised.
The model also identifies “indirectly affected” workers as those workers in localities where the prevailing minimum wage is less than the proposed minimum wage, and whose wages are greater than or equal to the proposed minimum wage but less than 115 percent of the proposed minimum wage. The model uses this cutoff point per findings by Dube, Giuliano, and Leonard (2015), who observe minimum wage spillover or “ripple” effects for workers earning up to 15 percent above newly implemented minimum wages. It is possible that spillover increases from a minimum wage change could carry further up the wage distribution. Wicks-Lim (2006) finds that past minimum wage increases in the United States between 1983 and 2002 affected workers at wage levels earning 123 percent of the new minimum wage. However, Wicks-Lim acknowledges that larger minimum wage increases may have less of a ripple effect as employers seek to limit the policy change’s impact on their overall labor costs. Given that many minimum wage increases being considered in recent years are larger than those that were typical between 1983 and 2002, it seems appropriate to set the upper bound at a lower level, although 115 percent of the minimum wage may still be a conservative assumption.
The model also accounts for proposed changes in the minimum wage for tipped workers—also known as the tipped minimum wage. As explained by Allegretto and Cooper (2014), under federal law and in most states, workers who regularly and customarily receive tips—such as restaurant servers, bartenders, nail salon workers, and casino workers, among others—can be paid a base wage less than the regular minimum wage if their weekly tips are sufficient to raise their effective average hourly wage rate for that week up to the regular minimum wage. Under federal law, this base wage can be as low as $2.13 per hour, again provided that over the course of the workweek, a worker’s tips make up the difference between $2.13 and $7.25 per hour (the regular federal minimum wage).
Tipped workers are identified in the data as those workers in customarily tipped occupations who are likely to be subject to the tipped minimum wage (or to be paid a base wage between the tipped minimum wage and the regular minimum wage). The occupations considered tipped, and their corresponding Standard Occupational Classification (SOC) codes, are listed below.
- 4040 – Bartenders
- 4060 – Counter attendants, cafeteria, food concession, and coffee shop
- 4110 – Waiters and waitresses
- 4130 – Dining room and cafeteria attendants and bartender helpers
- 4400 – Gaming services workers
- 4500 – Barbers
- 4510 – Hairdressers, hairstylists, and cosmetologists
- 4520 – Miscellaneous personal appearance workers
- 4120 – Nonrestaurant food servers in these industries (denoted by Census industry code):
- 8580 – Bowling centers
- 8590 – Other amusement, gambling, and recreation industries
- 8660 – Traveler accommodation
- 8670 – Recreational vehicle parks and camps, and rooming/boarding houses
- 8680 – Restaurants and other food services
- 8690 – Drinking places, alcoholic beverages
- 8970 – Barber shops
- 8980 – Beauty salons
- 8990 – Nail salons and other personal care services
- 9090 – Other personal services
There are other occupations in which workers often receive tips—such as massage therapists and taxi drivers—although it is less clear that workers in these occupations are paid a base wage less than the regular minimum wage. Thus, for the purposes of identifying workers likely to be impacted by changes in tipped minimum wages, the EPI model does not treat these occupations as tipped occupations.
The CPS-ORG and ACS wage data used in the EPI model do not allow users to identify the base wage paid to tipped workers, exclusive of tips. Hourly wage rates for tipped workers are calculated as their reported weekly earnings (inclusive of tips) divided by reported weekly hours of work. Consequently, it is difficult to distinctly identify directly versus indirectly affected tipped workers. Our approach is to delineate affected tipped workers as follows: Directly affected tipped workers are identified in the same way as workers directly affected by the change in the standard minimum wage. Any tipped worker with a reported hourly wage (inclusive of tips) below the proposed minimum wage is directly affected.
All other tipped workers with wages above the proposed minimum wage are considered indirectly affected if the proposed tipped minimum wage is above the prevailing tipped minimum wage in their locality. For example, even if a tipped worker reports hourly earnings of $16 per hour, there is no way to know whether that worker is being paid a base wage of the tipped minimum wage or something higher. Thus, the model considers these workers to be indirectly affected so that the resulting estimates describe the broadest possible workforce that would be affected by the proposed change in the tipped minimum wage.
For each step in the simulation, once the model has identified all directly and indirectly affected workers, it then makes a series of calculations to estimate the likely impact of the proposed minimum wage increase on those workers’ wages. Predicting the exact increase in affected workers’ wages is difficult, since the change in the minimum wage will affect workers with wages at a range of rates, from those right at the existing minimum wage (down to 80 percent of the existing minimum wage in the data) up to some point above the new minimum wage (set at 115 percent in the model). It is unlikely that all of these workers will receive a raise equal to the change in the minimum wage, and it is also unlikely that anyone with wages below the new minimum wage will have their wages raised only up to the new minimum.
Thus, to estimate wage impacts the EPI model makes two assumptions to smooth out the distribution of wages resulting from the simulated minimum wage increase: 1) All affected workers who are not tipped workers receive a raise of at least one-fourth the difference between the cutoff point for spillover effects (115 percent of the new minimum wage) and workers’ pre-policy-change hourly wage; and 2) all nontipped affected workers’ resulting hourly wage (post-policy-change) must be at or above the new minimum wage, except in cases where this would result in a raise larger than the change in the minimum wage. In these cases, workers’ raises are set equal to the difference between the new minimum wage and the existing minimum wage. The result of this method is that most directly affected workers are brought up to the new minimum wage. Those whose existing hourly wages were between the existing minimum wage and 80 percent of the existing minimum wage are given a raise equal to the change in the minimum wage. All indirectly affected workers receive a raise equal to one-fourth the difference between their existing wage and the spillover cutoff point of 115 percent of the new minimum wage.
For example, consider three workers under a proposed increase in the minimum wage from $7.25 to $8.55. In this scenario, the cutoff point for spillover effects would be 115 percent of $8.55, or $9.83. Prior to the policy change, worker A is paid $7.50 per hour, worker B is paid $8.60 per hour, and worker C is actually paid $7.25 an hour but misreports his wages as being only $7.00 per hour. In the EPI model, these workers’ raises would be calculated as follows:
- Worker A is directly affected. His raise equals the greater of (¼ × [$9.83 − $7.50]) = $0.58 or ($8.55 − $7.50 = $1.05) = $1.05. His resulting hourly wage is $8.55.
- Worker B is indirectly affected. Her raise equals the greater of (¼ × [$9.83 − $8.60] = $0.31) or ($8.55 − $8.60) = -$0.05 = $0.31. Her resulting hourly wage is $8.91.
- Worker C is directly affected. His raise equals the greater of (¼ × [$9.83 − $7.00]) = $0.71 or ($8.55 − $7.00) = $1.55. However, $1.55 is larger than the proposed change in the minimum wage. Thus, worker C’s raise is set at the change in the minimum wage, $1.30. His resulting hourly wage is $8.30.
For tipped workers, raises are calculated differently since it is not possible to know how a tipped worker’s existing base wage compared with the proposed tipped minimum wage and how much of a worker’s reported hourly wage is tips versus base wage. Thus, the model takes the following approach: For directly affected tipped workers, their raise is set equal to the change in the tipped minimum wage. For indirectly affected tipped workers, their raise is set equal to half the change in the tipped minimum wage. By applying half the increase in the tipped minimum wage, the model essentially assumes that some indirectly affected tipped workers receive more than the tipped minimum wage as a base wage, and some do not.
For all affected workers, the model estimates the resulting changes in their hourly wages, their usual weekly wages, and their annual earnings, assuming that they work year-round. With these values, the model calculates the difference between the wages that would result from the proposed minimum wage and the wages that would exist had there been no change in the minimum wage.
After recording these calculations, the model iterates to the next step, if any, in the proposed set of minimum wage increases. As was done prior to the first step, person weights are again adjusted to reflect the predicted population growth between the first and second steps in the proposed minimum wage increase. Estimated raises resulting from the policy change are applied to the wage values of all directly and indirectly affected workers; however, the simulation also maintains a counterfactual estimate of those workers’ wages—i.e., an estimate of those workers’ wages had there been no change in the minimum wage.
Wage values for all other workers are again adjusted to reflect natural nominal wage growth. However, in between the first proposed minimum wage change and any subsequent changes, the rate of “natural” nominal wage growth is assumed to be equal to the Congressional Budget Office’s projections for inflation as measured by the CPI-U, plus 0.5 percent (CBO 2018). In other words, the model assumes all workers not affected by the minimum wage increase experience real (inflation-adjusted) wage growth of 0.5 percent.
Prior to evaluating the impact of the next proposed increase, the model checks for any scheduled or expected changes in the prevailing minimum wage that would occur between the previous simulated increase and the next one, and adjusts wage values of affected workers as described for scheduled minimum wage changes between the data year and the month of the first simulated increase.
The same method for identifying directly affected workers, identifying indirectly affected workers, and calculating wage impacts is used for all subsequent steps.
In all EPI publications, the reported demographic characteristics of workers affected by the proposed minimum wage changes reflect the population affected through the final step of the proposal, as some workers who would be indirectly affected in earlier steps may become directly affected in later steps. Similarly, reported wage impacts describe the cumulative change in wages through the final step, relative to the wages that would exist at that same time in the counterfactual scenario in which no minimum wage change had occurred.
References
Allegretto, Sylvia A., and David Cooper. 2014. Twenty-Three Years and Still Waiting for Change: Why It’s Time to Give Tipped Workers the Regular Minimum Wage. Economic Policy Institute Briefing Paper no. 379, July 2014.
Bureau of Labor Statistics (BLS). 2015. “Labor Force Projections to 2024: The Labor Force Is Growing, but Slowly.” Monthly Labor Review, December 2015.
Congressional Budget Office (CBO). 2014. The Effects of a Minimum-Wage Increase on Employment and Family Income. February 2014.
Congressional Budget Office (CBO). 2018. The Budget and Economic Outlook: 2018 to 2028. August 2018.
Cooper, David. 2019. Raising the Federal Minimum Wage to $15 by 2024 Would Lift Pay for Nearly 40 Million Workers. Economic Policy Institute, February 2019.
Dube, Arindrajit, Laura Giuliano, and Jonathan Leonard. 2015. “Fairness and Frictions: The Impact of Unequal Raises on Quit Behavior.” IZA Discussion Paper no. 9149, June 2015.
Economic Policy Institute. 2019. The Impact of Raising the Minimum Wage to $15 by 2024, by Congressional District [interactive map].
Missouri Census Data Center (MCDC). 2018. Geocorr 2018 [data set]. http://mcdc.missouri.edu/applications/geocorr2018.html.
Ruggles, Steven, Sarah Flood, Ronald Goeken, Josiah Grover, Erin Meyer, Jose Pacas, and Matthew Sobek. IPUMS USA: Version 8.0 [data set]. Minneapolis, Minn.: IPUMS, 2018. https://doi.org/10.18128/D010.V8.0.
U.S. Census Bureau, 2013–2017 American Community Survey 5-Year Estimates. Accessed January 2019 at https://www.census.gov/acs/www/data/data-tables-and-tools/american-factfinder/.
U.S. Census Bureau, Current Population Survey Outgoing Rotation Group microdata (U.S. Census Bureau CPS-ORG). Various years. Survey conducted by the Bureau of the Census for the Bureau of Labor Statistics [machine-readable microdata file]. Accessed November 2018 at https://thedataweb.rm.census.gov/ftp/cps_ftp.html.
Wicks-Lim, Jeannette. 2006. “Mandated Wage Floors and the Wage Structure: New Estimates of the Ripple Effects of Minimum Wage Laws.” Political Economy Research Institute at the University of Massachusetts Amherst, Working Paper no. 116, May 2006.
