# Wage-age profiles now and then

The red line is the wage-age profile from 1990 to 2005 and the blue line is the wage-age profile from 1968 to 1980. I just picked those years randomly. These profiles were calculated in such a way that what you’re seeing is average “within person” wage profiles over their lives ((I estimated this equation $\inline \log w_{i,t} = \{D_a\}_{a=19}^{64} \beta+\gamma_i+\tau_t+\epsilon_{i,t}$ with 18 year olds the comparison group. The y-axis are the estimated two-way panel coefficient on the age dummy plus the average log wage for 18-year olds in the appropriate time period. Everything’s waaaay significant. These are heads of households with positive wages in the PSID. R code and the data set is available but its too big for me to post my hosted account so email me if you want it.)). This means there’s no funny business with changes in demographics or whatever:

In the good ol’ days, workers ramped up their wages early in their careers and then wages flattened out for the rest of their careers. In these evil dark ages of widening inequality, it takes longer for workers to get to their wages to peak and the peak is higher than before. Also, the peak comes so late there’s never a period of stagnant growth in their wages.

## 6 thoughts on “Wage-age profiles now and then”

1. Gabriel says:

Unless you’re also willing to include a short course in panel estimation, I don’t think I can use the data 😉 For some reason, panel data gets covered here in a dedicated 2nd year course. The same goes for time series

2. Step 1: put in “panel” data (records with individual id and time as variables)
Step 2: push panel estimation button (which gives you panel robust s.e.)
Step 3: get panel estimates out and interpret them as you would normal estimates (e.g. take derivatives)

If you need the theory (and why would you?), a chapter in Cameron’s microeconometrics is good. Panel estimates are just normal estimates but the data has been transformed in a clever way to get rid of individual and time fixed effects. Think of these fixed effects as individual and time dummy variables.

Panel’s are great because you don’t have to worry about finding control variables for fixed characteristics of individuals (e.g. education) or time (e.g. macro variables like unemployment).

3. Gabriel says:

At some point I will have to look into that. Summer plans are a bit different though.

P.S. Stop plugging books by your professors, LOL.

4. Dan in Euroland says:

Will,

I think you might be interested in this post on crooked timber.

5. heh… just posted on it before even seeing your comment. so yeah, I’m interested in it!

6. Mike says:

Interesting. Can you do this for median as well? Is the data set public?