Combining mlabel margins and mlabel p values in one coefplot

[Anouk Klootwijk]

Dear Statalists,

My question is about combining 2 types of mlabels in one coefplot in Stata 15.

I created a coefplot that displays margins of a logistic regression with the following command:

. coefplot, xline(0) mlabel format(%9.2g) mlabposition(12) mlabgap(*2)

I would like to add the pvalues in the same coefplot in the following manner:

. coefplot, xline(0) mlabposition(2) mlabel(cond(@pval<.001, "***", cond(@pval<.01, "**", cond(@pval<.05, "*", "")))) note("* p < .05, ** p <.01, *** p <.001")

When I put both mlabel and mlabel(cond(@pval<.001, "***", cond(@pval<.01, "**", cond(@pval<.05, "*", "")))) note("* p < .05, ** p <.01, *** p <.001") in the same command, the p values are not displayed.

How can I combine the two mlabels in one coefplot?

Thanks a lot in advance!

Regards, Anouk

[Andrew Musau]

coefplot is from Stata Journal. Please note the advice to identify community contributed commands and to present reproducible examples (FAQ Advice #12). The default specifying the option -mlabel- is to show the coefficient values next to their respective markers. Otherwise, -mlabel- option as in twoway scatter accepts a variable as input, which provides scope for you to program what you want. The following code is adapted for Stata 15, but a more efficient way to extract matrices is available in Stata 16 and later versions.

version 15
webuse lbw, clear
probit low age lwt smoke i.race
preserve
mat rtable= r(table)
mat res= (r(table)["b", 1...]\r(table)["pvalue", 1...])'
svmat res
drop if res1==0
format res1 %9.3f
gen label= cond(res2<0.001, string(res1)+"***", ///
cond(inrange(res2, 0.001, 0.0499), string(res1)+"**", ///
cond(inrange(res2, 0.005, 0.0999), string(res1)+"*", string(res1))))
coefplot, xline(0) mlabposition(2) mlabel(label) scheme(s1color)
restore

[Anouk Klootwijk]

Dear Andrew,

Thanks for your help. I just found out that my version of Stata is 15.1
When I run your code: mat res= (r(table)["b", 1...]\r(table)["pvalue", 1...])' an error invalid syntax r(198) appears.

Is there another way to achieve your coefplot with coefficients & pvalues for stata 15.1?

Thank you!

[Andrew Musau]

Sorry for the confusion. I created matrix rtable to avoid using r(table), so change that line to

mat rtable= r(table)
mat res= (rtable["b", 1...]\rtable["pvalue", 1...])'

EDIT: I am not even sure that the above syntax will work in Stata 15, and version control clearly does not work from #2, so you could also try

mat rtable= r(table)
mat res= (rtable[rownumb(rtable, "b"), 1...]\rtable[rownumb(rtable, "pvalue"), 1...])'

[Anouk Klootwijk]

Thanks a lot, the first code works already!!

How could I add in the code to wrap up the coefficients to three decimals ?

[Andrew Musau]

For strings, you can specify the display format as below

gen label= cond(res2<0.001, string(res1,"%9.3f")+"***", ///
cond(inrange(res2, 0.001, 0.0499), string(res1,"%9.3f")+"**", ///
cond(inrange(res2, 0.005, 0.0999), string(res1,"%9.3f")+"*", string(res1,"%9.3f"))))
coefplot, xline(0) mlabposition(2) mlabel(label) scheme(s1color)

[Anouk Klootwijk]

Thanks a lot Andrew, my coefplot is now exactly the way I wanted it to be!

[Anouk Klootwijk]

Is it also possible to combine multiple coefplots with corresponding mlabels in one figure?

I've generated labels for every coefplot seperately.

Thanks in advance!

[Andrew Musau]

Yes. The only requirement is that observations of the variable holding the labels should be ordered following coefficients in the combined coefplot. See Ben's method here to combine multiple models or using a matrix here.

[Anouk Klootwijk]

Thanks for your reply. In Ben's method there are two coefficients in the same graph. I tried your code as well for the matrix, but there was an error of invalid syntax in:
mat R= nullmat(R)\((r(table)["b", "weight"])\ (r(table)["ll".."ul", "weight"]))'

I think I mean a slightly different combination of coefplots, here is an example attached.Coefplots.gph

I tried to combine some codes and ended up with this figure, although some coefficients & p-values are missing. How can I get a combination of coefplots like this with all coefficients & p-values displayed?

Thanks in advance!

[Anouk Klootwijk]

When I add another subgraph in the coefplot, even more coefficients (mlabels) are missing. It seems like the total number of displayed mlabels corresponds with the number of independent variables...

[Andrew Musau]

I explain the principle here. If you struggle, provide a data example.

[Anouk Klootwijk]

I'm still struggling, so here is a data example.

webuse lbw, clear

* creating label for logistic regression on: 'low' odds ratios & p values

logistic low age lwt race

estimates store low

mat rtable= r(table)

mat res = (rtable["b", 1...]\rtable["pvalue", 1...])'

svmat res

drop if res1 == 0

format res1 %9.3f

gen label_low= cond(res2<0.001, string(res1,"%9.3f")+"***", cond(inrange(res2, 0.001, 0.0099), string(res1,"%9.3f")+"**", cond(inrange(res2, 0.01, 0.0499), string(res1,"%9.3f")+"*", string(res1,"%9.3f"))))

* creating label for logistic regression on: 'ui' odds ratios & p values

logistic ui age lwt race

mat rtable= r(table)

mat res_ui = (rtable["b", 1...]\rtable["pvalue", 1...])'

svmat res_ui

drop if res_ui1 == 0

format res_ui1 %9.3f

gen label_ui = cond(res2<0.001, string(res_ui1,"%9.3f")+"***", cond(inrange(res_ui2, 0.001, 0.0099), string(res_ui1,"%9.3f")+"**", cond(inrange(res_ui2, 0.01, 0.0499), string(res_ui1,"%9.3f")+"*", string(res_ui1,"%9.3f"))))

* store results for coefplot

logistic low age lwt race

estimates store low

logistic ui age lwt race

estimates store ui

coefplot low, mlabel(label_low) || ui, mlabel(label_ui)

I'm struggling with two things:

1. My coefplot looks like this (attachment), with missing mlabels for the odds ratios of race & _cons. Also, the mlabels in ui, are the ones that belong to low. When I plot the coefplots seperately for low & ui, all mlabels are present. Something is going wrong when I plot multiple graphs...

How can I put the corrects mlabels to the corresponding subgraphs?


2. For example in ui, odds ratio for weight at last menstrual period = 2.767, although in the coefplot it looks on the x-axis like it is -0.2... How can I put the odds ratios in the right position in the graph?


Thanks in advance!

[Andrew Musau]

In the future, please upload graph attachments in .png format.You can append this to the code in #13

preserve
keep label_low label_ui order
drop if label_low=="." & label_ui=="."
expand 2, gen(n)
replace label_low= label_ui if n
sort order n
coefplot low, mlabel(label_low) || ui, mlabel(label_low)
restore

All I did was to order the labels 11223344 corresponding to 1st coefficient left graph, 1st coefficient right graph, 2nd coefficient left graph, 2nd coefficient right graph, and so on.


ADDED IN EDIT: Had not noticed #2 in your post in #13. It seems to me that one unit in the axis scale represents 2 units in the coefplot graph. The equation is: New \(x\) = 2*Old \(x\) + 1. Therefore, you can change the labels as follows:

coefplot low, mlabel(label_low) || ui, mlabel(label_low) xlab(-2 "-3"  0 "1"  2 "5"  4 "9")

Res.:

[Anouk Klootwijk]

Hi Andrew,

Thanks a lot for your explanation, this works!! I removed the command order, because the error 'variable order not found r(111)' appeared.

How should I transform the code for a third coefplot?

For example

webuse lbw, clear

* create third label for ht
. logistic ht age lwt race
mat rtable=r(table)
mat res_ht = (rtable["b", 1...]\rtable["pvalue", 1...])'
svmat res_ht
drop if res_ht1 == 0
format res_ht1 %9.3f
gen label_ht = cond(res2<0.001, string(res_ht1,"%9.3f")+"***", cond(inrange(res_ht2, 0.001, 0.0099), string(res_ht1,"%9.3f")+"**", cond(inrange(res_ht2, 0.01, 0.0499), string(res_ht1,"%9.3f")+"*",string(res_ht1,"%9.3f" ))))

* store results ht
logistic ht age lwt race
estimate store ht

I tried the following:

keep label_low label_ui label_ht
drop if label_low=="." & label_ui=="." & label_ht=="."
expand 3, gen(n)

The n consists of 0 & 1, I think I need 0, 1 and 2 to be able to replace the labels for three graphs...
What command can I use ?

Thanks in advance!