Photo by Wolfgang Hasselmann on Unsplash
This post is following of above post.
From the previous post, NONNRGMAT has correlated to r_capi: squared rooted per capita gdp. Let's do regression analysys using R.
p-value for r_capi is almost 0. For TIME variables are not significant. Let's check if there are jointly significant. I use car::linearHypothesis() function.
I used matchCoef() function to make my Null Hypothesis. Then, I use linearHypothesis() function. The result shows p-value is 0.9628. So, TIME is not jointly significant.
Then, let's check if residuals is Homoskedasticiy or Heteroskedasticiy.
I use bptest() finction in lmtest package.
p-value is almost 0. So, this regression model has Heteroskedasticiy. Let's confirm lm() function and resid() function.
p-value is 2.689e-06, it is almost 0. So, we reject Null Hypothesis: residual is Homoskedasticity.
So, I compute Heteroskedasticity robust standard error with lmtest::coeftest( , cvov = hccm) function.
The conclusion does not change. l_gdp is significant at 5% level, r_capi is is significant at 0.11% level or below.
Next, let's do Weighted Least Square estimation. I use "weight = 1/r_capi" in lm() function.
Let's compare OLS estimation(lm_1) and WLS estimation(lm_wls) with statgazer package.
(1) is OLS and (2) is WLS. There is not significant change between OLS and WLS for r_capi coefficient.
That's it. Thank you!
Next post it
To read from the 1st post,