When to use logged / not logged variables

[Carys Wright]

Hi All,

I am carrying out research for my bachelor thesis looking at the effect of gin consumption on health outcomes, using a regional-level panel data set across time. I am using two measures of health for robustness: ARD (alcohol related deaths) and BADSAH (bad self-assessed health).

Controls:

TEH - control for Total Expenditure of Healthcare (£ billion spent in the specific region on Healthcare)
GDHI - control for income (Gross disposable household income in the region - £)
dbbinge - control for binge drinking behaviour


I am unsure whether to use the logs of TEH and GDHI - when I do I get extremely different results (different direction of relationship and varying significance). Does it make a difference? I've carried out my whole study using not the logs and written it all up - and now am apprehensive I may have needed to log the variables.

Many thanks,
Carys Wright

[Nick Cox]

Plot your data to see what is happening. Added variable plots do help. As earlier advised, size variables and rate variables may not mix well. Reversals of sign may relate to a bunch of extreme values.

[Carys Wright]

Hi Nick,

thank you for your reply. How do I plot my data may as I ask?

Carys

[Carys Wright]

Also when you say size and rate variables --> are these all not size variables?

[Nick Cox]

Added variable plots is a precise term you can search for.

[Carys Wright]

Hi Nick,

Okay thank you have looked that up. If the values are fairly closely distributed around the line of best fit, suggesting there isn't large extremity - does this mean that not using log isn't a big issue? I have attached the image as a png.

Many thanks,


Carys