Penn World Table Bleg

I need some help from the growth empiricists out there.  If you aren't one of them, stop reading.  Continuing will be a waste of your time.

For researchers studying economic growth, one of the standard resources for cross-country data has been the Penn World Table.  My 1992 paper with David Romer and David Weil (my most cited paper by a large margin) used this resource, as have numerous other papers in this literature.  In my intermediate macro book, I present a couple of figures presenting some of these data.

Here's the problem: It seems that the data have changed substantially in the most recent revision, and I cannot figure out why.

My intermediate macro text shows a scatterplot of per capita income and the investment share of GDP.  These two variables are strongly positively correlated.  When revising this figure with the newest data, I found that the correlation declines substantially (though is still positive).  When I looked into the source of the change, I found that the historical estimates of the investment share of GDP have changed, in some some cases by a lot.

Let me give you an example.  Take the investment share for Ghana in the year 2000.  According to version 6.2 of the data, the investment share was about 5 percent.  In version 7.0, it was about 21 percent.  This is one of the larger changes I have found, but it is not the only country for which there are sizable changes in the reported investment share of GDP.

I understand that the changes may be related to new information about the relative price of investment goods.  But the changes seem too large to be explained so easily, although perhaps I am wrong about this.  If anyone can shed light on the matter, I would be greatly appreciative.  Send me an email if you can help.
-----

Update: I have not yet fully figured this out, but readers have sent me some useful links.  If you are interested, click here, here, and here.