Help on regression

[Clem Zer]

Hello, I am new here an seeking help for my dissertation work.

I would like to test the value of 9 developmental factors on economic growth.

My dependent variable is: delta of GDP per capita in 2015-2004

My independent variables are:
Delta between 2015-2004 of primary education ratio
Delta between 2015-2004 in secondary education ratio
Delta between 2015-2004 in rural population (% in total population)
Delta between 2015-2004 in oil sells (% of GDP)
Delta between 2015-2004 in life expectancy
Delta between 2015-2004 in health expansion (%of GDP invest into health infrastructures)
Delta between 2015-2004 in Fertility rate

By Delta between 2015-2004 for each variable, I mean that I took the value in 2004 and subtracted it with the value in 2015, thus using the evolution of each factor to show the evolution of the GDP per capita, between 2004 and 2015.
I am using low and lower-middle income economies.

However, my findings are very very odd: most of my variables have a negative impact on GDP per capita (and most of them should have a positive effect on it) + most of my variables are statistically insignificant.

I am unfortunately lacking numerical and stata skills and this research work accounts for a large part of my degree.
Anyone with ideas of why my regression isn't making sense, it would be very helpful. Thank you in advance

Best

[Clyde Schechter]

This is a very broad question and there are a large number of possible answers. Among the possibilities to consider, in no particular order:

1. Are you sure your data are correct in your Stata data set? Has it been accurately transmitted into your Stata data set from the original sources? And are the original sources reliable: it is not at all unheard of for governmental or international sources of data to misreport statistics, either due to poor availability of information, or sometimes to intentionally distort for political/ideological reasons.

2. It is a bit odd to do this by just looking at two time points. Normally one would get more frequent data, say, annual, and look at trends over time. With only two time points, some atypical results in one or a few countries in one or both of those two years could be distorting results. Have you looked for outliers, not just in the raw variables but in their differences as well? Are they influential?

3. You show neither an example of your data nor a regression command, making it impossible for anybody to comment on whether you have done the analysis correctly.

4. Is your data set's collection of countries the full census of countries you wish to generalize about? Or do you have only a sample of the countries. If it was a sample, but not randomly selected, then the results may well be biased. Even if randomly selected, if it was subsequently trimmed down due to missing data values, it may not, in the end, be a properly random sample.

5. Are you sure your expectations about the directions of these associations are correct, and well supported by theory? Is there any reason that these particular countries, or these two years, might be ones to which the theory is not applicable?

6. As for your results not achieving statistical significance, the usual considerations apply: perhaps your sample is too small to detect realistic effect sizes. Perhaps your measurements are too noisy to enable sufficiently precise estimates of the effects. Perhaps there is a fair amount of multi-colinearity among the predictors you are using so that they are jointly significant, but not singly so.

7. Perhaps you have used a linear model but the relationships are non-linear and you are unable to capture them in your model. Have you explored your data graphically? What does theory say about this?