Calculate the standard deviation of a variable expressed in logarithms

[Ricardo Maciel]

Hi.

I have to calculate the standard deviation of a prediction that is expressed in logarithms in a regression with panel data,
the model form is log-log, I have the dependent variable "y" and the independent variables "x",
when I perform a regression of the linear form I can use the command "predict ystdp, stdp",
but when the model is of the log-log form the command "predict ystdp, stdp" calculates the standard deviation for the dependent variable "Lny"
but what I need is the standard deviation of the variable "y" for the log-log model, would there be some command in stata to perform this calculation?

Thanks for the help

[Nick Cox]

Why do you think you need this? If you want intervals for the predictions, you could exponentiate those for the logged response, or use a log link with xtgee.

[Weiwen Ng]

I suspect you meant standard error. The standard error is the standard deviation of a sample mean or some other quantity estimated from a model. The standard deviation is merely a descriptor of the amount of variance of some quantity. The syntax for -predict- refers to standard errors, not SDs.

The rest of my response would be identical to Nick's: if you really do need the SE of the predicted ys, you may be better off using a GLM or a GEE with a log link, which would not require you to transform the ys at all (although you still need to transform the xs if justified).

[Carlo Lazzaro]

Ricardo:
I do agree with the previous comments.
If you are really interested in ln_stanadard deviation, the painful procedure to calculate it is reported in http://journals.sagepub.com/doi/pdf/...2280202sm290ra.

[Ricardo Maciel]

Thanks everyone for answering, you're right Weiwen need the standard error (SE of the predictions). I'm working for a model that allows me to compare the electrical energy losses (variable "y") of each region of my country through a given complexity index by the prediction of the variable "y" for the last year of observation of each region (the log-log model gave the most significant coefficients), once the complexity index of each region was obtained, I obtained a ranking of ordered complexity from highest to lowest, I need the standard error to calculate the probability that a B region is positioned on top of a region A in the ranking, this probability would be to define the loss reduction goal of each region by means of the weighted average of the current energy losses of a region and that of the benchmark region, would there be an practice example to calculate with Stata the standard error of the variable y in a log-log model? Thanks for your time