I have a question that may be more of a statistics issue than a stata problem. I am using stata 11.1 for windows.
I am writing my thesis in which I have district level panel data from 625 districts from two surveys; 2001 and 2011. I want to use a first difference approach.
In the regression equations, one of the control variables I want to include is an indicator for wealth. The survey I am using does not explicitly ask questions on income, therefore I want to construct an index for wealth using asset ownership. There has been an abundance of studies and tutorials on this topic and from this literature I concluded the most suitable way of doing this for me would be through a Principal Component Analysis; since the observations I have, are on a district level and these are expressed in percentages instead of categorical values.
The problem I am running into is the construction of an index for two time periods and whether PCA is the appropriate methodology.
Since I will be comparing the indices from 2001 and 2011 by subtracting 2001 from 2011 (first difference), my first intuition was running two different principal component analyses; one for each time period. However this means there would be different principal components/eigenvectors for each time period.
Pca var_2001
Pca var_2011
In turn this would result in different weighting of the ownership of different assets in the final index for each year. As a result this would mean the two indices are not comparable.
A possible alternative would be doing one analysis for the data from both time periods, intuitively I would think the two indices now are comparable since they use the same weightings; the same principal components. However I am unsure whether this would be statistically justified and whether or not some time dependency would be lost since variables are time dependent.
pca var_2001 var_2011
I would love to hear your feedback on what would be the best way to construct a wealth index for my purpose.
I am writing my thesis in which I have district level panel data from 625 districts from two surveys; 2001 and 2011. I want to use a first difference approach.
In the regression equations, one of the control variables I want to include is an indicator for wealth. The survey I am using does not explicitly ask questions on income, therefore I want to construct an index for wealth using asset ownership. There has been an abundance of studies and tutorials on this topic and from this literature I concluded the most suitable way of doing this for me would be through a Principal Component Analysis; since the observations I have, are on a district level and these are expressed in percentages instead of categorical values.
The problem I am running into is the construction of an index for two time periods and whether PCA is the appropriate methodology.
Since I will be comparing the indices from 2001 and 2011 by subtracting 2001 from 2011 (first difference), my first intuition was running two different principal component analyses; one for each time period. However this means there would be different principal components/eigenvectors for each time period.
Pca var_2001
Pca var_2011
In turn this would result in different weighting of the ownership of different assets in the final index for each year. As a result this would mean the two indices are not comparable.
A possible alternative would be doing one analysis for the data from both time periods, intuitively I would think the two indices now are comparable since they use the same weightings; the same principal components. However I am unsure whether this would be statistically justified and whether or not some time dependency would be lost since variables are time dependent.
pca var_2001 var_2011
I would love to hear your feedback on what would be the best way to construct a wealth index for my purpose.
