Struggling to understand why P values are high in a panel data fixed effects model?

[Tatenda Smith]

I ran this fixed effects panel data regression with time dummys for Human Development Index against a mix of log and linear variables:

xtreg HDI lngdp iws phealth lnwbaid lnbilataid lnadbaid i.year, fe cluster (country_id)

Log GDP per capita, Improved Water Access %, Log Gross World Bank Aid received per capita, log DAC Aid received per capita, log Africa Development Bank aid received per capita and time dummys for each year



Before I used logs, my residuals for the currently logged variables were skewed, but now they are random. But for some reason, my P values are generally quite high and I'm not sure why?

[Carlo Lazzaro]

Tatenda:
two hypotheses:
1) your dataset has a (too) limited within-panel variation (and there's nothng yu can do about that, as data are what they are);
2) you have highly correlated variables (0.948 and 0.844 sound a tad suspect).
As an aside, you have coded a linear-log regression, which is not that frequent: do you have methodological reasons to follow this road?

[Tatenda Smith]

Tatenda:
two hypotheses:
1) your dataset has a (too) limited within-panel variation (and there's nothng yu can do about that, as data are what they are);
2) you have highly correlated variables (0.948 and 0.844 sound a tad suspect).
As an aside, you have coded a linear-log regression, which is not that frequent: do you have methodological reasons to follow this road?

Hi yes! Do you know any recommended tests for correlated variables?

I chose this type of regression because although I would have considered a log log model, HDI values are between 0-1 and the natural log would output a negative number and I was informed this would be bad for my model. I didn't want to log IWS as it's already a percentage, so the natural log I wouldn't be sure how to infer.

[Carlo Lazzaro]

Tatenda:
go -estat vce, corr-.