Well on further inspection of scatter graphs with the relevant independent variable the other author had in common ln(R&D) behaved in a more linear way than R&D so I guess I have my answer
I'm currently doing a little econometric analysis using Eviews. I'm using R&D spending (as a % of GDP) as the dependent variable, and basically just doing a cross country, 1 period regression with some other variables concerning institutions and such. I've seen the R&D variable used in a similar paper (with other explanatory variables), except the author used the (natural) logarithm. A teacher of mine advised me to just take the normal values if it was a percentage of income, but the author of the paper I mentioned didn't and I've been browsing around to see why he would do that, but unfortunately I've not been able to come with a satisfying answer. Taking the Ln or not does influence my results so I would like to hear some thoughts from all you fine lads here as to why one would take the logarithm of 'X as % of Y' in the first place. Some further info: Just doing a OLS, below are some descriptive statistics:
R_D Log(R_D)
Mean 1.131679 -0.295124
Median 0.769882 -0.261518
Maximum 3.708730 1.310689
Minimum 0.025416 -3.672369
Std. Dev. 0.939672 1.042697
Skewness 0.999514 -0.754034
Kurtosis 3.122134 3.570058
Jarque-Bera 9.191939 5.956573
Probability 0.010092 0.050880
Sum 62.24235 -16.23181
Sum Sq. Dev. 47.68106 58.70976
Observations 55 55
(Sorry, the above table doesn't come out well, first number belongs to normal value of R&D, second to the Log version)