GLM and Margins to obtain mean costs

[Dikshyanta Rana]

Hi everyone,

I am analysing a data with the aim to generate mean cost for two treatment arms A and B. I have 2 variables: type (Before or After) and age. Treatment A has 104 patients and B has 116 patients.

To do this, I used the following command:

bysort treatment: glm totalcost i.type i.age

This gave me exactly what I was looking for i.e. effect of type and age on total cost according to treatment A and treatment B.

To get the mean cost, I used the margins command.

margins type

This gave me the following table:

Predictive margins
Model VCE: OIM

Expression: Predicted mean total, predict () Number of obs = 116

Type	Margin	SE	z	P|z|	[95% Conf. Interval]
Before	41758.05	5360.599	7.79	0.000	31251.47	52264.63
After	64857.69	7530.048	8.61	0.000	50099.06	79616.31

I noticed that 'margins type' gave me the mean costs for both the types but only for treatment B, as shown by the number of observations.

So, I was wondering if you can help me figure out the reason behind this and how to obtain a similar table for treatment A?

[Carlo Lazzaro]

Dear Dikshyanta:
what you experienced can be explained considering that -margins- is fed by the second -glm- estimate only (ie, B, in your case).
To avoid this you should type:

glm totalcost i.type i.age if treatment=="A"
margins
glm totalcost i.type i.age if treatment=="B"
margins

As an aside, I suspect that a -glm- with -family(gamma)- and -link(log)- would be better for your data.

[Dikshyanta Rana]

Hi Carlo,

Thank you so much. You're absolutely right! It does provide a better fit. I forgot to add family and log while writing the question above.

Have a nice day!

Kind regards,
Dikshyanta