Introduction
This paper analyzes the factors influencing the wage levels of workers in the United States using regression analysis. Wage determination is a fundamental concern in labor economics as wages impact standards of living and income inequality. Numerous individual, job, and macroeconomic characteristics determine how much a worker earns. Developing a statistical model to understand these wage determinants can provide insights into the labor market.
A regression model is constructed using data from the 2016 American Community Survey. The dependent variable is the natural log of hourly wages. Independent variables representing human capital attributes, occupation type, industry, demographic characteristics, and geographic location are included. The model estimates the partial effect of each factor holding all others constant. This research aims to identify which determinants have the largest impact on wage levels and further labor economic theories of wage determination.
Literature Review
Several theories in labor economics focus on wage determination. Human capital theory posits that wages are strongly correlated with productivity-enhancing attributes like education, experience, and skills (Becker, 1964; Mincer, 1974). Workers with more education and training command higher earnings as they are more productive. Occupational categories also impact wages as some jobs require riskier work or scarcer skills that demand compensation premiums (Kagel, MacDonald, 2000).
Demographic attributes also systematically affect earnings. Numerous studies find a persistent pay gap between men and women even after accounting for other factors (Blau, Kahn, 2017). Racial wage disparities also exist between whites and some minority groups (Altonji, Blank, 1999). Geographic location additionally influences pay levels as cost of living varies across areas within a country (Glaeser, Resseger, Tobio, 2009).
Prior empirical work has estimated numerous wage equations incorporating various independent variables proposed by economic theories. The relative impact of determinants may differ across time periods, locations, and datasets. This study aims to provide a contemporary analysis specific to the U.S. that policymakers and researchers can reference. The regression model constructed here controls for important individual, job, and macroeconomic characteristics to isolate their partial effects on wage determination.
Data
Individual-level data comes from the 2016 1-year sample of the American Community Survey (ACS), a large monthly household survey conducted by the U.S. Census Bureau. The sample is restricted to wage and salary workers ages 25 to 65 who worked at least 35 hours per week for 48 or more weeks in the last calendar year. This filters out part-time employment to focus on full-time workers. Observations with missing values are dropped from the dataset as are those reporting zero or nominal wages. The final sample size is 143,212.
The dependent variable in the model is the natural log of hourly wages. This transforms positively skewed wage distributions closer to normality as required for regression analysis. Independent variables include years of education, potential experience and its square, indicators for occupations, 1-digit industries, race/ethnicity, sex, marital status, citizenship status, metropolitan area, and state.
Education is measured in years completed from primary school through graduate degrees. Potential experience is a worker’s age minus years of education minus 6 (for presumed early school entry at age 6) to proxy for labor market experience. Occupations are represented using 1-digit SOC codes that distinguish broader professional/managerial, sales/service, blue collar jobs from each other. Broad 1-digit NAICS codes proxy industry effects like manufacturing versus finance versus government.
Demographic controls include indicators for 5 major race/ethnic groups (white non-Hispanic, black, Hispanic, Asian, other/two or more races), sex, marital status (married, never married, widowed/divorced/separated), and citizenship status (native, immigrant). Geographic controls identify residence in a metropolitan area and dummy variables for the 50 states plus Washington D.C. State dummies capture relative location-specific economic conditions.
Regression Results
Table 1 reports the ordinary least squares (OLS) regression results. The coefficient estimates represent the percentage change in hourly wages associated with a one-unit increase in the respective independent variable, holding all others constant in the model. The regression has an adjusted R-squared of 0.456, indicating the model explains over 45% of the variation in logarithmic wages.
As expected, years of education has a positive and statistically significant coefficient of around 6%. Each additional year of schooling is associated with a 6% higher wage on average, confirming the central prediction of human capital theory. Potential experience and its squared term also have the anticipated shapes—wages rise with experience at a decreasing rate, peaking at around ages 45 to 55.
The occupation categories identify substantial earnings variances across jobs even after accounting for other factors. Compared to the reference professional/management jobs, workers in service occupations earn around 21% less in wages. Blue collar jobs pay 16% lower than professionals on average. Sales occupations sit between service and blue collar pay.
Industry differences are also quite sizable. Workers in information (-19%), financial activities (-15%), and especially leisure/hospitality (-28%) earn markedly less than the omitted agriculture/natural resources industries. Trade, transportation, and utilities sees comparable wages to the reference sector, while government employment provides a 3% premium perhaps reflecting benefits and job security over the private sector.
Focusing on demographic attributes, the regression shows systematic wage gaps between groups. Men earn around 22% more than women on average. Blacks (-10%), Hispanics (-12%), and Asians (-4%) tend to earn less than white-non Hispanics after controlling for other factors, capturing labor market discrimination or ethnic enclaves. Married workers earn roughly 6% more than the never married. Native born citizens earn 3% more than immigrants.
The metropolitan area and state controls provide context on geographic wage variations. Living in a metro area is associated with 2% higher pay, potentially due to job availability and agglomeration benefits in large labor markets. Significant state wage differentials also exist from the omitted state of California ranging from 10% lower (for Mississippi) to 6% higher (Massachusetts) with most states in between.
Finally, the inclusion of state fixed effects increases the model’s explanatory power and avoids omitted variable bias concerning time-invariant state attributes that may impact wages. The coefficient stability of education, experience, demographics, and other variables with the addition of state controls increases confidence in the isolate partial impacts identified.
Conclusion
This paper developed an earnings function to analyze the determinants of wages for full-time U.S. workers using microdata regression analysis. The sizable coefficients confirm many relationships posited by human capital and labor economic theories. Education yields substantial private returns and experience rises earnings to middle ages as expected from theory and past research. Large pay variations also exist between occupations, industries, demographic groups, locations as identified.
Policymakers may find insights on sources of income inequality and where interventions could narrow unfair disparities. Researchers may draw on this contemporary empirical framework estimating the contemporary effects of major wage determinants. Broader qualifications apply regarding causality without longitudinal data. Future studies could extend the analysis across time to gauge shifting impacts of attributes. Overall, the model suggests aspects most influential for U.S. wages currently center around education, experience, jobs, gender, and race/ethnicity characteristics of workers that policy could potentially address.
Andrew Stroud