Panel-data exploratory data analysis

Question

I have a dataset for around 40k firms over fiscal years 1950-2011 with about 430k firm-years. If I'm not mistaken I have panel-data. In addition, the firms are nested within 9 industries.

I created a unique identifier ticn for each firm. Years are indicated by fyear. For now my variables of interest are yearly sales sale, yearly advertising xad, and yearly R&D expenses xrd. I have industry dummies indicated by sicagg. I am interested in the relationship between yearly sales and advertising/R&D expenditures, including some control variables.

Currently I am in the exploratory phase of my research.

So my objective, I want to get a feel for the data, give descriptives and maybe make a few plots.

First I computed between and within descriptive statistics (mean, stdev, min and max). I also made scatter plots between sales, R&D and advertising. In addition, I plotted the time-series of the yearly average advertising expenses for each industry in a nice graph.

Can you give me ideas for additional analyses? Thanks in advance!

Answer 1

I always start by doing a PCA (Principal Component Analysis) in R because it takes almost no writing. Say you have all this in a data.frame that we call data.

pca <- prcomp(data)
# Screeplot.
plot(pca)
# Biplot.
biplot(pca)

For R users, there is also the ggplot2 library. I know that it can do wonders for data representation, but I don't know how to use it. Maybe someone will suggest something with it?

Answer 2

It's not clear to me what you've graphed when you say "scatter plots between sales, R&D, and advertising". For example, have you done something like:

library (lattice)
xyplot (sale ~ xrd | year, groups=sicagg)
xyplot (sale ~ xrd | sicagg, groups=year)

Not sure what sicagg is; I assume it's a factor variable in my example.

In your plot of advertising per industry, did you plot lines for the averages and points for the specifics, coded by industry? Density plots might also be useful:

densityplot (~sale, groups=xrd)
densityplot (~xad, groups=xrd)

etc. Once you get complex and combine graph types, lattice gets complicated fairly fast, but it makes these kinds of plots easy.