After weeks of googling myself, I am pleased to announce that my working paper on human capital augmented production functions is finally live on RePEc. You can view it here. You might recognize the theoretical proof from a previous post I wrote on here a few months ago while the paper was still in development.

What is in the paper that was not in the blog post are as follows:

- A reasonable review of the literature
- A simulation to drive the point home
- Zombie puns

**UPDATE**: In the interest of transparency and what I meant to do but just forgot (thanks to Robert Vienneau for reminding me), here are my R scripts. I apologize in advance about the run time of the second one:

Humbug SimulationIf you have any trouble with the scripts, make sure you clear your R environment from one before running the other.

Success Simulation

1st para: "...these economists see no problem defining human capital for the sake of empirics..." Are you aware of this Ian Steedman paper? It might be usefully referenced here.

ReplyDeleteFootnote 2: I assume the last sentence is talking about price and real Wicksell effects. Is it referring to anything else?

I'm not sure what I think about reporting a T stat of zero and a p value of unity. Presumably, Caselli reports an F-stat for his ratio of variances. Can you get a F stat from your UNDEAD simulated data.

I like the Shaikh paper in which he analyses actual data and data simulated from a non-neoclassical model (with shifting Leontief production functions, if I remember correctly). He finds that the empirical methods used by Solow and followers fit his non-neoclassical data better than the real data. Do you plan on doing something like this or expanding your description of your simulated data?

Thanks for your feedback!

DeleteI'm actually not familiar with the Steedman paper. Thanks for the reference. I'll definitely check it out.

You're correct on the interpretation of the footnote.

I mean, I could do an F-stat test here, but to me it seems like a distraction. The point is that Caselli's method of calculating the value of A is identically equal to the modified Shaikhian version (except for rounding error). Hence what I'm reporting is that the null hypothesis that the difference between the two equals zero is for all intents ant purposes "true". (I know how statisticians have rhetorical positivist hang-ups about reporting true null hypotheses. Maybe I should change my blog title to "Vulgar Statistics")

That's what I did basically did with the second test, but I used a random data process for robustness. The data reported with figure 4 uses a thousands of random aggregate income samples of random sizes and uses the methods to calculate Caselli's success score with his method (equation 5) and the reverse (equation 16) using the modified Shaikhian method to calculate A. Again, I take the difference (histogram in figure 4 - note that R uses a 16-digit float) and calculate the t for Ho: differences = 0.

I want to put my R scripts up somewhere so folks can run this data themselves, but I have no idea where or how to do that. Any suggestions would be helpful.

Again, thank you so much for your feedback. I really respect your work, so this means a lot.

Okay, so I'm an idiot and totally forgot that I already store my scripts on Dropbox. I've updated this post with links to my scripts.

Delete