Healthcare Systems Idea?

[paul-opheim]

Here is that healthcare systems classification paper I found:

https://www.econstor.eu/bitstream/10419/64809/1/726547131.pdf

I'm guessing this idea will not be ideal, both because of the categorical variable issue and because there are two categories (private health system and social-based mixed type) that only have one country within them. I suppose we could redefine the question to be National Health Services vs Etatist Social Health Insurance, or something like that.

• [ ] Try dimensionality reduction on the fixed effects too? But then again, this doesn't make too much sense because why would there be measurement error in the fixed effects.

[ijyliu]

@marionoro did you see where the data is? I think we would need it to be a panel, because otherwise a single observation for 30 countries is not much to go on

[ijyliu]

This seems like a better data source: https://data.oecd.org/healthres/health-spending.htm

We can use the government/compulsory percent of total health spending. And that will also be a continuous variable.

I bet this has been reposted on the world bank indicators site where there might also be more countries i guess

[ijyliu]

See this memo: https://github.com/ijyliu/ECMA-31330-Project/blob/main/Release/PCR_and_Measurement_Error.pdf

[ijyliu]

We probably need more covariates. However, these need to be selected as ones we think there will be measurement error in...

Also maybe add fixed effects

[ijyliu]

I added fixed effects: https://github.com/ijyliu/ECMA-31330-Project/blob/main/Release/PCR_and_Measurement_Error.pdf

But as you can see, the coefficients using the covariates versus PC1 don't really change much.

I think we would need to add many more covariates. But I'm wary of doing this because then we basically start including things that aren't really 'mismeasured' because they don't really represent the same concept.

[ijyliu]

I just added more covariates, we have 11 now. But also I feel like I'm kind of stretching the World Bank's usefulness https://github.com/ijyliu/ECMA-31330-Project/blob/main/Release/PCR_and_Measurement_Error.pdf

[ijyliu]

Try dimensionality reduction on the fixed effects too?

But then again, this doesn't make too much sense because why would there be measurement error in the fixed effects?

[ijyliu]

Ok, so now with more covariates there's kind of a difference in the coeffcients, particularly in the case without fixed effects: https://github.com/ijyliu/ECMA-31330-Project/blob/main/Release/PCR_and_Measurement_Error.pdf