Dear Statalists,
I was just looking over a project of a friend of mine , and in one of his regressions he includes population density and total population. At first I thought: perfect multicollineartiy. Then I realized that while total population and population density can be expressed as a linear relationship for one country, that the linear relationship varies across countries. Still, Im wondering if specifications like this are problematic? The data set contains 15 countries across 28 years. My intuition tells me that due to the perfect correlation of the two variables per country, a lot explanatory power is lost respectivley more variation is needed to identify the signal even though the correlation across all observations corresponds only to 0.15. Furthermore, I'd say that roughly instead of having 15*28 observations to separate the effects of both variables from each other, I basically only have the variation of 15 observations as the variation of population density and total population can not be separated per country. My fellow phd students did not agree on this, but could not tell me where my supposedly error of thought lies, maybe you can.
Even though this is not a stata specific problem, I still hope to get an answer.
Kuba
I was just looking over a project of a friend of mine , and in one of his regressions he includes population density and total population. At first I thought: perfect multicollineartiy. Then I realized that while total population and population density can be expressed as a linear relationship for one country, that the linear relationship varies across countries. Still, Im wondering if specifications like this are problematic? The data set contains 15 countries across 28 years. My intuition tells me that due to the perfect correlation of the two variables per country, a lot explanatory power is lost respectivley more variation is needed to identify the signal even though the correlation across all observations corresponds only to 0.15. Furthermore, I'd say that roughly instead of having 15*28 observations to separate the effects of both variables from each other, I basically only have the variation of 15 observations as the variation of population density and total population can not be separated per country. My fellow phd students did not agree on this, but could not tell me where my supposedly error of thought lies, maybe you can.
Even though this is not a stata specific problem, I still hope to get an answer.
Kuba
