Regional Wage Growth During the Recession

Nominal wages at the aggregate level have been found to rise during the recent recession (or at least not decline).This is surprising given that aggregate labor conditions were so weak.Many economists (myself included) have viewed this as a puzzle.The goal in this question is to assess whether this is a puzzle do to the composition of workers changing over time.In particular, I want you to do the following.

1.  Use data from the 2000-2016 March CPS (downloaded from IPUMS CPS web site) to examine the time series (annual) trend in both nominal and real wages for 25-54 year old men not living in group quarters. (You need to download the group quarters variable). Wages will be defined as annual earnings (last year) divided by annual hours (last year). Annual hours can be computed by multiplying weeks worked last year * usual hours worked (either last year or currently). You should do this for a sample of all workers with a positive wage last year. You can convert to real wages using any deflator you wish (I use the June CPI-U in my work). June is in the middle of the year – that is why I pick it.

a.  The first part of the assignment is to plot mean and median wages over time for this sample. How have wages evolved over the last 15 years in the U.S. What is the time series correlation between wage growth and changes in the employment to population ratio (defined within the same sample) between 2000-2015 (note that wages measured using the 2016 CPS sample actually refers to earnings in 2015). All means should be weighted by survey weights provided by the CPS.

b.  In the second part of the question, I want to demographically adjust the data to adjust for composition changes over time (based on observables). As employment rates fell, they fell more for low educated workers than high educated workers. That means the average wages you measure in part (a) will actually be comparing different types of workers in 2010 relative to 2006. To demographically adjust the data, create age-skill groups in each year. Use 5 year age ranges (25-29, 30-34, etc.) and 5 education groupings (less than high school, high school only, some college, bachelor’s degree, more than a bachelor’s degree). Compute wages within these cells for each year (again weighting the data using survey weights). However, when computing time series trends, we will fix the population weights for each cell at year 2000 levels.[1] So, for each year, you will compute a measure of wages by multiplying the means within each cell during that year by the 2000 shares of population in each cell (and then sum across cells within the year). By doing this, you will be able to compute a demographically adjusted wage series during the 2000 – 2015 period. How do these demographically adjusted wage series compare to your series in part (a). Discuss the differences. Do wages look more or less responsive with respect to changes in employment?

c.  In the third part of the question, I want you to impute wages for those not working. So your sample will now expand to all workers – even those without a measured wage. We will do the imputation in a crude way. Using this new sample, define demographic groups by age-skill for each year. For each person without out a wage in a given year – match them to their demographic cell in that year. For your imputation, assign those with a missing wage in that year to the 33rd percentile of the wages within that cell (for that year). Note – those with the highest wages will be at the 99th percentile. This assumes that those with a missing wage (of a given age-skill type) are disproportionately drawn from the bottom part of their age-skill wage distribution. Still adjust the wages for changing demographic composition by fixing the weights for each demographic cell at their 2000 level (using the full sample including the zeros). How do the time series patterns in these demographically adjusted wages (including the zeros) compare to the wage series created in part b. How would things change if you used the 50th percentile to conduct the imputation?

2.  Redo parts (a)-(c) using data from the 2000 Census and the 2001-2014 American Community Survey (downloaded from the IPUMS USA website). You should use the same sample restrictions as above. The group quarters restriction is much more important for the ACS sample. You should be aware that the educational codes change in the ACS in 2007 or 2008 – so, you will need to adjust your code to account for this.

Do the raw wage patterns in the CPS and ACS line up with each other? Do the demographic adjustments make a difference? Does imputing the wages make a difference?

When writing up your results, make sure you take time to describe your figures and flush out your conclusions. Both the write up and figures should be self-contained. That means, if someone just looked at your figures, they should be able to understand what is going on (axis labeled, title, lines labeled, a detailed note describing the figure, etc.). Also, someone should be able to read your write up and figure out what you are trying to convey. Your write up does not have to be long – but, it should be clear.

[1] I.e. the share of the population that lives in each group.