Over at PostEverything, but here’s the figure–five wage/comp series ably smashed together by my colleague Ben Spielberg using principal components analysis (a useful way to avoid cherry-picking the series that tells the story you like, PC analysis pulls out the common, underlying trend on the combined series).
Source: BLS, see data note in WaPo post.
The figure above is a weighted average of year-over-year wage growth for five different data series:
–Employment cost index: hourly compensation
–Employment cost index: hourly wages
–Productivity series: hourly compensation
–Median weekly earnings: full-time workers
–Average hourly earnings: production, non-supervisory workers
The data all come from the BLS and are non-seasonally adjusted, except for the productivity series.
To derive the figure, we:
–take yearly changes in the data (e.g., q1/q1/, q2/q2, etc.) and run a principal components analysis on the yearly changes.
–using the first principal component, we divide each coefficient by the standard deviation of the series that corresponds to that coefficient.
–obtain weights by dividing the resultant value for each series by the sum of all the resultant values.
–multiply the matrix of series data by a vector of weights from the previous step to obtain the plot.
All of the above is necessary just to generate a series of percent changes that is a weighted average of the underlying data. But this plot just shifts the scale of the first PC series that any statistical software package will generate (i.e., it is perfectly correlated with that series (r=1)).
For further info or for an E-views program that automates the above, write to Ben Spielberg (firstname.lastname@example.org). We thank Jesse Rothstein for helping with this scaling transformation.
Data in Excel are here.