Plotting regression coefficients

[Hale Isaac]

Hello,

I’m running a quantile regression using David Powell’s -qregpd- package (SSC) on the 25^th, 50^th and 75^th percentiles of my outcome wb. I am trying to graph the coefficients in a way similar to what coefplot (SSC) does (attached image). I have tried adjusting the code suggested here by Clyde Schechter , but can’t get it right.

forvalues i=0.25(0.5)0.75{
qregpd WB COVID2 COVID3 COVID4 econ2 econ3 econ4 food2 food3 emp2 emp3 income2 income3 age mar2 mar3 educ2 educ3 educ4 urban2 under_six2 safety_net2, id(Findid) fix(wave) optimize(mcmc) noisy draws(3000) burn(300) arate(.5)
}


capture postutil clear
tempfile holding
postfile handle quantile coefficient using `holding'

forvalues i=0.25(0.5)0.75{
    qregpd WB COVID2 COVID3 COVID4 econ2 econ3 econ4 food2 food3 emp2 emp3 income2 income3 age mar2 mar3 educ2 educ3 educ4 urban2 under_six2 safety_net2, id(Findid) fix(wave) optimize(mcmc) noisy draws(3000) burn(300) arate(.5)
    post handle (`i') (_b[COVID2])
}
postclose handle

use `holding', clear
graph hbar coefficient quantile, over(WB)

The help file says the results are stored in the following way:

qregpd saves the following in e():
 
    Scalars        
      e(N)                number of observations
 
    Macros        
      e(cmd)              "qregpd"
      e(title)            "Quantile Regression for Panel Data"
      e(indepvars)         Righ-hand side variables
      e(depvar)            Left-hand side variable
      e(gamma)             Values on the gamma vector
 
    Matrices      
      e(b)                Parameter values
      e(V)                Variance-covariance matrix of parameters
      e(solutions)        Parameter values from grid search; may contain multiple solutions.
 
    Functions     
      e(sample)           marks estimation sample

[Jared Greathouse]

Just out of curiosity, why not just use coefplot?

[Hale Isaac]

First time I tried to do it, I got the error no estimates found or all dropped, but now it seems to have worked! Produced the graph I need. Feeling pretty silly, thanks for pointing out the obvious Jared.

[Jared Greathouse]

🤣 we all have moments like that, it just puzzled me when you said you knew about coefplot but didn't want to use it.

We've all been there before, or at least, I know I have.

[Tiago Pereira]

Have you tried -metan- to plot those estimates? I have used -metan- many times to plot medians (Q1, Q3). You loop over the predictors, collect medians, Q1s and Q3, and plot them using -metan- with the options ooverall nowt nostats.

gene equalweight = 10
metan median q1 q3, nooverall nowt nostats wgt(equalweight)

You can add labels, modify sticks and add information on the left- and right-hand sides using the options lcols and rcols.

All the best,

Tiago

[Jared Greathouse]

Tiago Pereira Using metan to report summary stats?

[Tiago Pereira]

Yes, in -metan-, we can plot anything that has a point estimate, a lower bound and an upper bound. They can be mean (95% CI), posterior medians (95% Crl) or median (Q1, Q3). Of course, the summary results will make no sense, that is why we have to use the option ooverall nowt nostats.

[Jared Greathouse]

Tiago Pereira I'd never even thought about it like that. So theoretically, it could be used to plot, say, a variety of average treatment effects from a variety of different estimators. I'd never even thought of that, but it seems like a great idea I'd love to experiment with. I'll give it a go.