Multiple Legends and Reducing Variables with Four Panes in Coefplot

[Ian Curtis]

Hey there Stata users,

I am trying to plot the results for a select number of independent variables from 8 separate regressions using coefplot. Among these 8 estimates, there are 4 dependent variables. For each of the 4 dependent variables, I am measuring the effect of 1 key independent variables (plus some controls).

I want to show the effects of each of these four outcomes with coefplot as a separate window pane. Because I will estimate two different independent variables on the same outcome, but in different models, it would be more efficient to include them in the same window. Therefore, I want two plots (from separate models) in the same window four times.

The figure I've made is legible but rough. The problem is 1.) I only have one legend that labels two of my independent variables. Ideally, I'd like to show a legend with all four possible independent variables. 2.) In every pane, all four possible independent variables are shown on the y axis. This is unnecessary. There should only be two. Using the "keep" or "drop" command only seems to work for the whole figure and I cannot figure out how to selectively use it for certain panes.

This is an impasse I can't work past. To make matters worse, my data is protected state administrative data on a server without internet access meaning I cannot reproduce it. I used the auto dataset to try and recreate my setup. Please forgive how crude this is.

You can see I estimate price and headroom first and alternate the independent variable (trunk and displacement). Then I estimate MPG and weight with rep78 and gear ratio as IVs. Each model controls for length. Again, I want a legend for trunk, displacement, rep78 and gear ratio. I also only want the relevant IVs shown (e.g. when estimating price, we would see trunk and displacement but NOT rep78 and gear ratio).

reg price trunk        length 
    eststo m1
reg price displacement length  
    eststo m2
    
reg headroom trunk        length  
    eststo m3
reg headroom displacement length  
    eststo m4

reg mpg rep78     length  
    eststo m5
reg mpg  gear_ratio   length 
    eststo m6

reg weight rep78  length
    eststo m7
reg weight gear_ratio length  
    eststo m8
    
coefplot     (m1, label("Trunk")) ///
            (m2, label("Displacement")), ///
                    bylabel(Price Estimates) ///
            ||     (m3, label("Trunk")) ///
                (m4, label("Displacement")), ///
                    bylabel(Headroom Estimates) ///
            ||     (m5, label ("Rep 78")) ///
                (m6, label ("Gear Ratio")), ///
                    bylabel(MPG Estimates) ///
            ||    (m7, label("Rep 78")) ///
                (m8, label ("Gear Ratio")), ///
                    bylabel(Weight Estimates) ///
                keep (trunk displacement rep78 gear_ratio) ///
                xline (0) byopts(title("I don't know anything about cars")) ///
                mlabel(cond(@pval<.001, "***", /// 
                cond(@pval<.01, "**", ///
                cond(@pval<.05, "*", ""))))

And here is how the figure looks.

[Andrew Musau]

coefplot is from Stata Journal, you are asked to explain. I don't think that a legend showing all 4 variables is any good here. Stata interprets your plot as a single graph rather than multiple graphs (notice that your confidence interval lines and markers have the same colors across subgraphs). This should be clear from the description of the -bylabel()- option.

bylabel(string) provides a label for the subgraph.

You could plot two subgraphs at a time and use the graph combine command, but there are better ways to present your results. See examples here.