Creating a forest plot with odds ratios

[Tara Boyle]

Hello I would like to create a forest plots for the odds ratios of the depedent variable on the outcome.
I have 5 models.

I have tried

//Mi  - model 1
quietly logit mi_30 asa [pw=_weight], or if procedure == 0
//stroke - model 2
queitly logit stroke_30 asa [pw=_weight], or if procedure == 0

// If function is not allowed

The plan was then to use -coef-plot

coefplot, drop(˙cons) xline(1) eform xtitle(Odds ratio)

I only wanted to show the outcomes of mi_30 and stroke_30 for procedure == 0 and procedure ==1 on the coef plot.
However, the -coef- will plot all the OR of all the predictors inlcuding ASA - which is not what I want

Is there an alternative?

[Andrew Musau]

coefplot is from SSC, as you are asked to explain (FAQ Advice #12).


EDITED:

I only wanted to show the outcomes of mi_30 and stroke_30 for procedure == 0 and procedure ==1 on the coef plot.

What do you mean by this? These are your outcomes. You can only plot estimated coefficients.

[Tara Boyle]

With regards to -if- function I was following the advice of:
https://www.stata.com/meeting/german.../de14_jann.pdf
slide 21


just wanted to plot the odds ratios of my procedure 0 and procedure 1 just to graphically depict the outcome

[Andrew Musau]

In the link, the coefficients on the regressors are plotted. "foreign" is just a grouping variable.

[David Fisher]

Returning first to your parallel thread, https://www.statalist.org/forums/for...or-odds-ratios:

It is somewhat misleading to claim that the command:

regress price mpg trunk length turn if foreign==0

... is the source for the plot that you show. From your link, the full sequence of commands run is as follows:

regress price mpg trunk length turn if foreign==0
estimates store D
regress price mpg trunk length turn if foreign==1
estimates store F

That is, two different regressions are run: one for "foreign" and one for "domestic". Their respective results are stored, and then displayed simultaneously on a single plot.

You are correct in your original assertion that "one can not present the OR of both base-level and value 1". Specifically, an OR is an odds ratio: in this case, the ratio of level 1 to level 0. There is no such thing as an OR for level 0 or level 1 alone (that would be an "odds", and would require a quite different interpretation).

Now, you say "Hello I would like to create a forest plots for the odds ratios of the dependent variable on the outcome" ... and then "I only wanted to show the outcomes of mi_30 and stroke_30 for procedure == 0 and procedure ==1 on the coef plot".
It is not clear what your dependent variable is; if it is not ASA (as you state), then presumably it is "procedure 0 vs 1". But you cannot have "procedure" as your dependent variable and then fit separate models for procedure==0 and procedure==1 (i.e. the grouping variable). That does not make sense. To return to the "regress" example: price is the outcome; mpg, trunk, length and turn are all dependent variables, and foreign is the grouping variable. You are currently seeking to have procedure do two jobs at once. (That is, for example, the jobs of both mpg and foreign in the "regress" example.)

[Tara Boyle]

David Fisher Thank you for pointing this out

I've managed to figure this out after some time (using sample dataset below), by using the following code

logit mi_30 treatment price, or
est store m1
logit uti treatment price, or
est store m2

coefplot m1 m2, sort nolabel drop(_cons) keep(*treatment)  //keep - to keep only OR for treatment for each outcome

I obtain the following plot:

However my question is: As I am going to have around 5 models, this may make it a bit complex for the reader, with 5 different colours.
Is there another alternative solution to perhaps try present the plot in this manner - I've tweaked the plot myself, not sure if this is possible however.
Really appreciate your expertise and time !

* Example generated by -dataex-. For more info, type help dataex
clear
input str18 make int price float gear_ratio byte foreign float(treatment mi_30 uti) byte(_est_m1 _est_m2)
"AMC Concord" 4099 3.58 0 1 1 1 1 1
"AMC Pacer" 4749 2.53 0 0 1 0 1 1
"AMC Spirit" 3799 3.08 0 0 0 0 1 1
"Buick Century" 4816 2.93 0 0 0 1 1 1
"Buick Electra" 7827 2.41 0 1 1 0 1 1
"Buick LeSabre" 5788 2.73 0 0 1 1 1 1
"Buick Opel" 4453 2.87 0 0 1 0 1 1
"Buick Regal" 5189 2.93 0 0 1 1 1 1
"Buick Riviera" 10372 2.93 0 1 1 1 1 1
"Buick Skylark" 4082 3.08 0 0 0 1 1 1
"Cad. Deville" 11385 2.28 0 1 1 0 1 1
"Cad. Eldorado" 14500 2.19 0 0 0 1 1 1
"Cad. Seville" 15906 2.24 0 1 1 1 1 1
"Chev. Chevette" 3299 2.93 0 1 0 0 1 1
"Chev. Impala" 5705 2.56 0 0 0 0 1 1
"Chev. Malibu" 4504 2.73 0 0 0 . 1 0
"Chev. Monte Carlo" 5104 2.73 0 0 0 0 1 1
"Chev. Monza" 3667 2.73 0 1 1 0 1 1
"Chev. Nova" 3955 2.56 0 0 1 1 1 1
"Dodge Colt" 3984 3.54 0 1 0 0 1 1
end
label values foreign origin
label def or

[David Fisher]

Hi Tara,

Further down the page of https://repec.sowi.unibe.ch/stata/co...g-started.html is an example of the "bycoefs" option, which I think is what you need?

clear
input str18 make int price float gear_ratio byte foreign float(treatment mi_30 uti) byte(_est_m1 _est_m2)
"AMC Concord" 4099 3.58 0 1 1 1 1 1
"AMC Pacer" 4749 2.53 0 0 1 0 1 1
"AMC Spirit" 3799 3.08 0 0 0 0 1 1
"Buick Century" 4816 2.93 0 0 0 1 1 1
"Buick Electra" 7827 2.41 0 1 1 0 1 1
"Buick LeSabre" 5788 2.73 0 0 1 1 1 1
"Buick Opel" 4453 2.87 0 0 1 0 1 1
"Buick Regal" 5189 2.93 0 0 1 1 1 1
"Buick Riviera" 10372 2.93 0 1 1 1 1 1
"Buick Skylark" 4082 3.08 0 0 0 1 1 1
"Cad. Deville" 11385 2.28 0 1 1 0 1 1
"Cad. Eldorado" 14500 2.19 0 0 0 1 1 1
"Cad. Seville" 15906 2.24 0 1 1 1 1 1
"Chev. Chevette" 3299 2.93 0 1 0 0 1 1
"Chev. Impala" 5705 2.56 0 0 0 0 1 1
"Chev. Malibu" 4504 2.73 0 0 0 . 1 0
"Chev. Monte Carlo" 5104 2.73 0 0 0 0 1 1
"Chev. Monza" 3667 2.73 0 1 1 0 1 1
"Chev. Nova" 3955 2.56 0 0 1 1 1 1
"Dodge Colt" 3984 3.54 0 1 0 0 1 1
end

set seed 1

gen byte outcome1 = round(uniform())
gen byte outcome2 = round(uniform())
gen byte outcome3 = round(uniform())

logistic outcome1 treatment price
estimates store outcome1
logistic outcome2 treatment price
estimates store outcome2
logistic outcome3 treatment price
estimates store outcome3

coefplot outcome1 || outcome2 || outcome3, drop(_cons price) bycoefs xtitle(Odds ratio of treatment) xlabel(-.693 "0.5" 0 "1" .693 "2" 1.386 "4")

[Tara Boyle]

Thanks, just downloaded schemepacks. Graph looks great.

An additional question, is there a way I can rename the 'outcome1', I tried

coeflabels(outcome1="Myocardial Infarction")

However this didn't work, I suppose as this is for coefvariables
I then tried the shortcut to rename my models as the variables I want to see in graph.

I managed to get away with most variables renamed by storing the models as -estimates store MI-

The only limitation is that a model name such as : VTE&PE is not accepted by stata

Question 1:
Do you have any suggesstions?

Question 2:
Do you have an idea if I can add the p-values; at outcome1 or is the only alternative to add as notes?