Subsitution analyses

[Nadine Khoury]

Hello, I wanted to see if someone can help me with my analysis.
I´m assessing the association between ultra-processed food consumption (in g/day) according to NOVA classification system and CVD risk factors.
We wanted to explore what happens when we substitute 100g of ultra-processed food (NOVA 4) with 100g of non-processed food (NOVA 1) in stata.
I know that we have to use a linear regression model and then use the lincom command by theory but i am having a hard time implementing it.
The idea is to remove 100 g from NOVA 4 group and add 100 g from NOVA 1 in our population, keeping the same g/d consumed by the participants (but making their intake healthier),How we can translate that to Stata syntax?

[George Ford]

say means are NOVA4 = 1000 and NOVA1 = 1000

margins, at(NOVA4 = (1000 900) NOVA1 = (1000 900)) post

it will give all 4 combinations and you can pick the one's you want. due to post, they'll be in e(b). you can also test for the difference.

[Tany Garciduenas]

George Ford: what will change in the syntaxis if the NOV4 is, at the same time, the exposure? and can we replace the numbers (1000 900) by just (100) which is the amount that we are looking to replace?

Thank you!

[George Ford]

It will only give you the value at 100, so you can't compare it to anything. If you want to evaluate how Y changes for a 100 unit change, you need two values of NOV 100 units apart, and probably near the mean if there's any non-linearity in the model. Or just make the calculations off the coefficients (though margins will ensure you got that correct).

from your #1, this might be better:

margins, at(NOVA4 = (1000 900) NOVA1 = (1000 1100)) post

[Tany Garciduenas]

Thank you very much, George Ford.
I have been trying to calculate the coefficients with margins as you commented but I'm not sure if it's correct. To my understanding, it would be:

margins, at (NOVA4=100 NOVA1=100) post

Thank you!