Coefplot with multiple models

[Jay Rege]

For my research I'm estimating the same regression with three different dependent variables, which are different variations of GDP per capita. In order to graphically compare my coefficients I'm using coefplot to plot them with a 95% confidence intervals above each other. So far I'm using the following code to store each estimation result as a seperate model to then tell coefplot to plot all three models.

reg logY_1 x1
estimates store results_1
reg logY_2 x1
estimates store results_2
reg logY_3 x1
estimates store results_3

coefplot results_1 results_2 results_3, drop(_cons)

This results in a graph with a single "x1" on my y-axis and three different lines in my graph, which represent my three different models as the legend shows.

In order for a neater graphical presentation I would like Stata to show my three model names (results_1, results_2 and results_3) on the y-axis (instead of the single "x1") at the height of the corresponding line in the graph, just as coefplot does for estimates with multiple independent variables.
So far I've been tweaking around with the legend options, but it doesn't trun out nicely.
I think this problem may be solved by storing all three results in a single matrix, but I have trouble how to construct such a matrix and how to tell coefplot to get it right.

[Rich Goldstein]

if I understand you correctly, use of the "swapnames" option on your -coefplot- command should give you what you want;

[Jay Rege]

Unfortunately not, I have tried the option "swapnames" before and it only results in changing the "x1" label on the y-axis to "_".

[Ben Jann]

Try:

sysuse auto
reg mpg foreign
estimates store results_1
reg trunk foreign
estimates store results_2
reg turn  foreign
estimates store results_3
coefplot (results_1 \ results_2 \ results_3), drop(_cons) aseq swapnames

Or with custom labels:

coefplot (results_1, aseq(whatever I want) ///
        \ results_2, aseq(something else) ///
        \ results_3, aseq(no more ideas)), drop(_cons) swapnames

Or:

coefplot (results_1 \ results_2 \ results_3), drop(_cons) aseq swapnames ///
    coeflabels(results_1 = "whatever I want" ///
               results_2 = "something else"  ///
               results_3 = "no more ideas")

ben

[Parth Aks]

Hi,

I am working with several models : 3 Outcome variables over 2 sub-groups and have 2 covariates. I want to plot to the sub-groups side by side and models should be separated by covariates, and within the covariates the I want to chose only a few outcomes e.g.
the three outcomes would be mpg, trunk, and turn; subgroups would be rep78=4 and rep78 = 5; and covariates are: foreign and gear_ratio.

For foreign (covariate), results from trunk and mpg regressions should be plotted for both subgroups, whereas for gear_ratio (covariate) results from turn and mpg should be plotted for both sub-groups. See ideal graph depicted below:



I modified the approach given above as follows but (1) it places foreign's coefficients on different lines for different subgroups whereas they should be on same horizontal line i.e. coefficients for foreign from trunk_s4 and trunk_s5 should appear on same horizontal line (2) I am unable to select different models across the two covariates.

sysuse auto, clear
reg trunk foreign gear_ratio if rep78 == 5  // Outcome = Trunk and sub-group = 4
estimates store trunk_s5
reg trunk foreign gear_ratio if rep78 == 4    // Outcome = Trunk and sub-group = 5
estimates store trunk_s4

reg mpg foreign gear_ratio if rep78 ==5  // Outcome = MPG and sub-group = 4
estimates store mpg_s5
reg mpg foreign gear_ratio if rep78 ==4  // Outcome = MPG and sub-group = 5
estimates store mpg_s4

reg turn foreign gear_ratio if rep78 ==5 // Outcome = Turn and sub-group = 4
estimates store turn_s5
reg turn foreign gear_ratio if rep78 ==4  // Outcome = Turn and sub-group = 5
estimates store turn_s4

coefplot (trunk_s4) (mpg_s4) || (turn_s5) (mpg_s5)  , drop(_cons) aseq swapnames

It produces the following graph which doesnt solve my issues listed above


I was able to solve problem 2 of using different models for each covariate and plot them across different sub-groups, using matrix command.

forvalues x = 4/5 {
foreach var in trunk mpg turn {
    est restore `var'_s`x'
    parmest, saving(temp_`var'_s`x'.dta, replace) eform
}
}
use temp_trunk_s4.dta, clear
gen str outcome = "trunk"
gen byte subgroup = 4
append using temp_trunk_s5
replace outcome = "trunk"
replace subgroup = 5 if subgroup == .
forvalues x = 4/5 {
foreach var in mpg turn {
    append using temp_`var'_s`x'.dta
    replace subgroup = `x' if subgroup ==.
    replace outcome = "`var'" if outcome ==""
    
}
}
sencode parm, gen(covno)
sencode outcome, gen(outno)
mkmat estimate min95 max95 if covno==1 & subgroup==4 & inlist(outno,1,2), matrix(sg4top)
mkmat estimate min95 max95 if covno==2 & subgroup==4 & inlist(outno,3,2), matrix(sg4bot)
mkmat estimate min95 max95 if covno==1 & subgroup==5 & inlist(outno,1,2), matrix(sg5top)
mkmat estimate min95 max95 if covno==1 & subgroup==5 & inlist(outno,3,2), matrix(sg5bot)
mat sg4 = sg4top\sg4bot
mat sg5 = sg5top\sg5bot
matrix rowname sg4 = "Trunk" "MPG" "Turn" "MPG2"
matrix rowname sg5 = "Trunk" "MPG" "Turn" "MPG2"

coefplot   matrix(sg4[,1]), ci((sg4[,2] sg4[,3])) bylabel("Rep78=4") ///
        || matrix(sg5[,1]), ci((sg5[,2] sg5[,3])) bylabel("Rep78=5") ///
        mlabel format(%9.2f) mlabposition(12) mlabgap(*2) mlabsize(vsmall) ///
        coeflabels(,labsize(small) wrap(20)) ///
        msymbol(smsquare) ciopts(recast(rcap)) scheme(s1mono)

But it produces the following graph: I am fine with this but it is unable to add sub-titles for Foreign or Gear-Ratio using this approach. Would appreciate if anyone can point how to add sub-titles for this matrix approach or modify the simpler approach using coefficients directly.


Thank you

[Pietro Biroli]

Dear Ben,
I have a related problem. is it possible to use the groups() or headings() option with swapnames?

Specifically, say that I want to run the main specification (results 1, 2, and 3) and then re-run then same regression for different sub-groups:

sysuse auto
*** Baseline
reg mpg foreign
estimates store results_1
reg trunk foreign
estimates store results_2
reg turn foreign
estimates store results_3

*** Group 4
reg mpg foreign if rep78 == 4
estimates store results_1_g4
reg trunk foreign  if rep78 == 4
estimates store results_2_g4
reg turn foreign  if rep78 == 4
estimates store results_3_g4

*** Group 5
reg mpg foreign if rep78 == 5
estimates store results_1_g5
reg trunk foreign  if rep78 == 5
estimates store results_2_g5
reg turn foreign  if rep78 == 5
estimates store results_3_g5

The following code produces a nice graph, but without any groups names:

coefplot ///
    (results_1, aseq(whatever I want) \ results_2, aseq(something else) \ results_3, aseq(no more ideas)) ///
    (results_1_g4, aseq(whatever I want) \ results_2_g4, aseq(something else) \ results_3_g4, aseq(no more ideas)) ///
    (results_1_g5, aseq(whatever I want) \ results_2_g5, aseq(something else) \ results_3_g5, aseq(no more ideas)) ///
, drop(_cons) swapnames ///
groups(result_1 result_2 result_3 = "Baseline"  ///
            result_1_g4 result_2_g4 result_3_g4 = "Group 4" ///
            result_1_g5 result_2_g5 result_3_g5 = "Group 5" ///
            )

Is there a way of solving this? thanks
Pietro

[Devorah Manekin]

Dear Pietro (or anyone else who may know),
Did you ever figure this out?
Thanks,
Devorah

[Ben Jann]

With respect to #5, I think the following should do:

sysuse auto, clear
reg trunk foreign gear_ratio if rep78 == 5  // Outcome = Trunk and sub-group = 4
estimates store trunk_s5
reg trunk foreign gear_ratio if rep78 == 4    // Outcome = Trunk and sub-group = 5
estimates store trunk_s4

reg mpg foreign gear_ratio if rep78 ==5  // Outcome = MPG and sub-group = 4
estimates store mpg_s5
reg mpg foreign gear_ratio if rep78 ==4  // Outcome = MPG and sub-group = 5
estimates store mpg_s4

reg turn foreign gear_ratio if rep78 ==5 // Outcome = Turn and sub-group = 4
estimates store turn_s5
reg turn foreign gear_ratio if rep78 ==4  // Outcome = Turn and sub-group = 5
estimates store turn_s4

coefplot (trunk_s4, keep(foreign) \ mpg_s4 \ turn_s4, keep(gear_ratio) \) ///
      || (trunk_s5, keep(foreign) \ mpg_s5 \ turn_s5, keep(gear_ratio) \) ///
      || , drop(_cons) aseq swapnames eqrename(*_s4 = "" *_s5 = "") ///
           order(trunk turn mpg) bylabels("Subgroup 4" "Supgroup 5") ///
           eqlabels(, asheading)

[Ben Jann]

With respect to #6, I am not sure what the desired graph is, but maybe the following helps:

sysuse auto
*** Baseline
reg mpg foreign
estimates store results_1
reg trunk foreign
estimates store results_2
reg turn foreign
estimates store results_3

*** Group 4
reg mpg foreign if rep78 == 4
estimates store results_1_g4
reg trunk foreign  if rep78 == 4
estimates store results_2_g4
reg turn foreign  if rep78 == 4
estimates store results_3_g4

*** Group 5
reg mpg foreign if rep78 == 5
estimates store results_1_g5
reg trunk foreign  if rep78 == 5
estimates store results_2_g5
reg turn foreign  if rep78 == 5
estimates store results_3_g5

coefplot (results_? results_?_g4 results_?_g5), drop(_cons) aseq swapnames ///
    groups(results_? = "Baseline" results_?_g4 = "Group 4" results_?_g5 = "Group 5") ///
    coeflabels(*1* = mpg *2* = trunk *3* = turn)

[amandacpac]

Ben, I am trying to do something very similar to what you are doing in post #8, though I have 29 different dependent variables (and 1 independent variable), and I am running all 29 models twice (over two different time points). However, when I run the example code you provided, I get the following plot:


What I would REALLY like to have happen is for the two trunk variables (trunk_s4 and trunk_s5) to be on the same line (i.e., side-by-side), AND to be able to rename that to "Trunk". And the same goes for MPG, Turn, etc. Any chance there is a way to do that??

Oh, and I would only need the "Foreign" part of this plot, as I only have the 1 independent variable.

Thanks!

[Jane Quan]

Hi all!

I have a few queries regarding the use of the Coefplot command and hope to receive assistance from this forum. My objective is to create a coefficient plot for a single variable within a set of regression models. The command I have used is as follows:

coefplot (r16, label(Korean Style Restaurants)) (r12, label(Chinese Style Restaurants)) (r10, label(Japanese Style Restaurants)) (r8, label(Western Style Restaurants)) (r3, label(Other Eastern and Western Style Restaurants (r2, label(Cafeterias)) (r13, label(Catering Food Services)) (r11, label(Confectioners Shops)) (r15, label(Pizza, Hamburger, Sandwich and Similar Food Services)) (r14, label(Chicken Shops)) (r6, label(Noodle Houses)) (r1, label(Other Lunch Counters n.e.c.)) (r9, label(General Amusement Drinking Places)) (r5, label(Dancing Amusement Drinking Places)) (r4, label(Other Drinking Places)) (r7, label(Non-Alcoholic Beverages Places)) , keep(logkorean)  xline(0) mlabel(cond(@pval<.001, "<0.001" + "***",cond(@pval<.01, string(@pval,"%9.3f") + "**", cond(@pval<.05, string(@pval,"%9.3f") + "*",string(@pval,"%9.3f"))))) scheme(s1color)

However, the resulting graph is displayed in a such way:

Is there a way to align the legend labels on the y-axis to correspond with each respective model?
Additionally, is it possible to group these labels based on their specific characteristics, like categorizing them under 'restaurant' and 'drinking places' and perhaps enclosing each group within large parentheses? Many thanks for your assistance!

[Jane Quan]

Hi all!

I have a few queries regarding the use of the Coefplot command and hope to receive assistance from this forum. My objective is to create a coefficient plot for a single variable within a set of regression models. The command I have used is as follows:

coefplot (r16, label(Korean Style Restaurants)) (r12, label(Chinese Style Restaurants)) (r10, label(Japanese Style Restaurants)) (r8, label(Western Style Restaurants)) (r3, label(Other Eastern and Western Style Restaurants (r2, label(Cafeterias)) (r13, label(Catering Food Services)) (r11, label(Confectioners Shops)) (r15, label(Pizza, Hamburger, Sandwich and Similar Food Services)) (r14, label(Chicken Shops)) (r6, label(Noodle Houses)) (r1, label(Other Lunch Counters n.e.c.)) (r9, label(General Amusement Drinking Places)) (r5, label(Dancing Amusement Drinking Places)) (r4, label(Other Drinking Places)) (r7, label(Non-Alcoholic Beverages Places)) , keep(logkorean) xline(0) mlabel(cond(@pval<.001, "<0.001" + "***",cond(@pval<.01, string(@pval,"%9.3f") + "**", cond(@pval<.05, string(@pval,"%9.3f") + "*",string(@pval,"%9.3f"))))) scheme(s1color)

However, the resulting graph is displayed in a such way: [ATTACH=CONFIG]n1733263[/ATTACH]

Is there a way to align the legend labels on the y-axis to correspond with each respective model?
Additionally, is it possible to group these labels based on their specific characteristics, like categorizing them under 'restaurant' and 'drinking places' and perhaps enclosing each group within large parentheses? Many thanks for your assistance!

I think I have solved the problem by using "ylabel" & "aseq swapnames" commands!