Plotting a fitted Line for log scale axis

[Andre Maier]

Hello,

I have to plot a simple fitted line for exports relative to GDP (the variable is called "norm_exports") on the y axis and bilateral distance (variable "km") on the x axis. However, I am having problems, since I have to depict the y axis in log scale. I know that I can therefore not use the standard twoway lfit command:

twoway (scatter norm_exports km)(lfit norm_exports km)

Therefore I tried to use predict for a regression of the log of my export variable on distance:

quietly reg log_norm_exports km

predict gr

label var gr "Linear prediction"

twoway (scatter norm_exports km)(line gr km, sort), yscale(log)

However, the code I used is not working. I would be really thankful if one could tell me where my mistake was.

Thanks a lot.

[Nick Cox]

Perhaps you need either

quietly reg log_norm_exports km
predict gr
* !!! or 10^gr if you used log10() to generate log_norm_exports 
replace gr = exp(gr) 
twoway (scatter norm_exports km)(line gr km, sort), yscale(log)

or

quietly poisson norm_exports km  
predict gr
twoway (scatter norm_exports km)(line gr km, sort), yscale(log)

-- noting that these are not exactly the same. Usually the second is a better idea. The fallacy in your code is that ysc(log) is fine for your data, but your predictions are already on log scale and not commensurate.