search

ijyliu / ECMA-31330-Project

Econometrics and Machine Learning Group Project
2 stars 1 forks source link

Bonhomme Notes #29

Closed paul-opheim closed 3 years ago

paul-opheim commented 3 years ago

"Prof. Bonhomme: make clear that the latent variable is uni dimensional. What if development is muti-dimensional? Use a combination of empirical application and simulation that mimic the empirical setting. Do you obtain different results if you just average the different measures as opposed to use PCA?"

ijyliu commented 3 years ago

That's an interesting idea for a comparison- averaging the ols coefficients

ijyliu commented 3 years ago

Or wait, maybe he means averaging the mismeasured variables. Which I guess would be kind of similar to doing the PCR except it's the optimal combo of the variables...

paul-opheim commented 3 years ago

I interpreted it as he meant the latter. Yeah it seems like it should be reasonably similar, but it would be interesting to compare the two methods.

ijyliu commented 3 years ago

make clear that the latent variable is uni dimensional. What if development is muti-dimensional?

Does this mean it's easily explained with a one/a few PCs? Unfortunately, this is the scree plot:

image

It seems to me that we need like 7+ PCs. So maybe we want to use more than one PC in our PCR.

I guess the issue is then that it's a little less fair to compare what we have to just a regression with the government health share and a single GDP measurement because then we would be leaving things out...

Linked to #15

ijyliu commented 3 years ago

resolved with #15, though we also have versions with more economic measures