When to use log variables in a regression?

[James Wilde]

Good Morning,

I have put together a panel data set of 24 countries over a period of 30 years looking at the impacts of various indicators on development, using the HDI as my measure for human development. These indicators are urbanisation (measured as urban population as % of total population), military expenditure (as a % of GDP), trade (as a % of GDP), inflation (price level of household consumption) and government spending (price level of govt. consumption).

A similar study to mine has taken the natural log of all the variables for their empirical analysis and I do not fully understand the reasoning for doing this. However, I do not want to miss out on doing this if it results in the empirics of my study being wrong due to this mistake. Any advice on this matter would be much appreciated.

Kind regards,

[Carlo Lazzaro]

James:
whenever I read queries like yours I wonder whether the original poster can rely on a teacher/supervisor/professor to give her/him some good guidance.
That said, in the study you mention Authors adopted a log-log regression model (see James H. Stock and Mark W. Watson. Introduction to Econometrics (4th Edition) Global Edition. Pearson Education, 2020: page 293-4).

[James Wilde]

Hello Carlo,
The reason for asking such a question on this forum is because my supervisor has not replied to me for a month, hence taking matters into my own hands.
Thank you for your advice, it is much appreciated.

[Carlo Lazzaro]

James:
actually I surmised something similar.
That said, the previous reference (and many others) can provide you with comprehensive replies to all your questions on regression.
In addition, please note that there are other issues to check in regression model, such as heteroskedasticity, serial correlation of the idiosyncratic error, endogeneity and the correctness of the functional form of the regressand, just to mention the most important.