# Avoiding spurious regression in meta-regression using stationary variables like GDP

#### samuelle

##### New Member
I have been reading everything I can get my hands on about spurious regression but can't seem to definitively find out what is best in regressing cross-sectional data including non-stationary variables.

Example data:

Code:
Country    GDPyearofstudy         Popdensyearofstudy   Year of study Total(2013$) UK 2462484285580 2.63 2011 0.5 Brazil 2143067871760 0.23 2010 1.5 USA 13095400000000 0.34 2005 2.3 USA 14958300000000 0.37 2010 1.5 Total observations: 49 I am conducting a meta-regression of values obtained from studies between 1990-2011. My y variable is Total (2013$) and my x variables include GDP and Population density to help understand the valuation. The year the study was undertaken is also included. I was told using GDP and Population density could skew results through being non-stationary, so I attempted to calculate GDP growth as 100*log(GDP/lagGDP). As I only have one observation for some countries however, this has mostly produced errors as there is no lag on a single observation.

My questions therefore are:

1. Do I use GDP and Population Density for the year the original study was taken, or
2013 which I have standardised Total(2013\$) to?

2. If this is cross-sectional, do I have to worry about GDP and Population Density
being non-stationary?

3. If they are non-stationary, will using the log of each be enough, or do I have to use GDP growth which in my case mostly fails to produce a result?

4. Can I simply adjust the standard errors?

I have run regressions using GDPyearofstudy; GDP2013; logGDPyearofstudy; logGDP2013; GDPgrowth and all fitted vs. residual plots look fine, Breusch-Pagan test is fine and all variables show a Pearson correlation of <0.6.

Thanks.