Marginsplot with catagorical variables on x axis - custom spacing

[John Porter]

I have a dataset where the outcome is binary (y), and a predictor variable is in the form of ordinal categories (x). These ordinal categories, for the data that I'm analysing, are numbered 2, 3, 4, 6, 7, and 8. There was no type 5 condition for these data.

I ran the logistical regression (logistic germinated i.s_treatment), and then margins (margins i.s_treatment) for the categories.

	Delta-method
Margin	Std. Err.	z	P>z	[95% Conf.	Interval]
s_treatment
2	.2158891	.0190877	11.31	0.000	.1784779	.2533003
3	.226945	.0194258	11.68	0.000	.1888712	.2650188
4	.2810673	.0198444	14.16	0.000	.2421729	.3199617
6	.3776728	.0224277	16.84	0.000	.3337153	.4216304
7	.2911443	.0210381	13.84	0.000	.2499104	.3323781
8	.3500235	.0206868	16.92	0.000	.3094782	.3905688

I then ran marginsplot.

Marginsplot includes a space where category 5 should be, treating it as if it were a continuous variable. I want to remove this gap, is there a simple way to do so?

[Andrew Musau]

egen s2_treatment= group(s_treatment), label

Then use s2_treatment in place of s_treatment.

[Scott Merryman]

An elaboration on Andrew's reply:

sysuse auto,clear
//Create gap
replace rep = rep+1 if rep > 3

egen temp_rep = group(rep), label
local lab: variable label rep
label variable temp_rep "`lab'"

qui reg price i.temp_rep
qui margins i.temp_rep
marginsplot, xlabel(1(1)3 4 "5" 5 "6")

[Bruce Weaver]

Thanks for that example, Scott. But note that the xlabel option seems to be unnecessary. Running the code below, I find that the g1 and g2 plots are identical.

qui reg price i.temp_rep
qui margins i.temp_rep
marginsplot
graph rename g1
marginsplot, xlabel(1(1)3 4 "5" 5 "6")
graph rename g2

Cheers,
Bruce

[Andrew Musau]

Thanks for that example, Scott. But note that the xlabel option seems to be unnecessary. Running the code below, I find that the g1 and g2 plots are identical.

As you are labeling the grouped variable (using the -label- option of egen), it will take the to be grouped variable's labels if it is labeled, otherwise it will take its values as labels. So, indeed, modifications to the x-axis labels are unnecessary.