Plotting regression coefficients as a histogram

[Iva Mihaylova]

Hello,

I have to run about 35 regressions and plot the estimates of the key regression coefficient on a histogram. Do you know how to do this?

I have started experimenting with the sample data provided below but I could not manage to plot the estimates of the key regression coefficient mpg on a histogram. First I thought that coefplot may be appropriate, but then I realised that it cannot nicely visualise so many estimates and somehow the data should be plotted as a histogram.



sysuse auto, clear

regress price mpg trunk length turn if foreign==0
estimates store ELF1

regress price mpg trunk length turn if for==1
estimates store ELF2

regress price mpg trunk length if for==1
estimates store ELF3

regress price mpg trunk length turn if foreign==0
estimates store ELF4

regress price mpg trunk length turn if foreign==0
estimates store ELF5

regress price mpg trunk length turn if for==1
estimates store ELF6

regress price mpg trunk length turn if foreign==0
estimates store ELF7

regress price mpg trunk length turn if for==1
estimates store ELF8

regress price mpg trunk length turn if foreign==0
estimates store ELF9

regress price mpg trunk length turn if for==1
estimates store ELF10

regress price mpg trunk length turn if foreign==0
estimates store ELF11

regress price mpg trunk length turn if for==1
estimates store ELF12

regress price mpg trunk length turn if foreign==0
estimates store ELF13

regress price mpg trunk length turn if for==1
estimates store ELF14

regress price mpg trunk length turn if foreign==0
estimates store ELF15

regress price mpg trunk length turn if for==1
estimates store ELF16

regress price mpg trunk length turn if foreign==0
estimates store ELF17

regress price mpg trunk length turn if for==1
estimates store ELF18

coefplot ELF1 ELF2 ELF3 ELF4 ELF5 ELF6 ELF7 ELF8 ELF9 ELF10 ELF11 ELF12 ELF13 ELF14 ELF15 ELF16 , xline(0) keep(mpg) drop(_cons)

[Nick Cox]

If your real problem is regression for the same response and predictors for subsets of the data, statsby is one way of producing a results set from which a histogram is immediate.

[Iva Mihaylova]

Thank you, Nick, this works for me.

[Nick Cox]

OK. The principle behind statsby is still one that you might well aim at. That is, put your estimates into variables because for histograms you need variables!

As I'm imagining that people have often wanted to do this, but also thinking that I can't at this moment recall a canned way to do it. I am going to watch to see if someone else can identify a wheel already invented/