[Help] Marginal effect on Stata

[James Gascoigne]

Hi. My question is from Stata (15.1), but also requires some statistical knowledge, which is why I am calling on you.
I'm doing a homework, and I have three different models.
1: ln (y) = b0 + b1 * X + error
2: y = b0 + b1 * ln (x) + error
3: ln (y) = b0 + b1 * ln (x) + error

I have a database with the variables x and y. I have to tell among the 3 models, which assumes a decreasing marginal relationship between Y and X.
How can I do with Stata?

When I use:

* 1: log-lin
gen ln_y = ln (x)
reg ln_y x
mfx
* 2: lin-log
gen ln_x = ln (x)
reg y ln_x
mfx
* 3: log
reg ln_y ln_x
mfx

At the mfx command, I always get postives results >0 (but I should have a result <0 in order to have a decreasing marginal relationship, isnt it ?).

Even when I derive the functions, I get positive results. Do you have an idea that can prevent from failing this assignement ?

[Phil Bromiley]

As the FAQ notes, we are not here to help with homework. It is completely inappropriate to present our work as your's in a course which is what you obviously intend. It also means you won't learn what the assignment was intended to teach. Please don't try this again.

[Richard Williams]

I won't do your homework but I will suggest some readings that might help. For margins, see the 5 handouts (or at least the first one) that are at

https://www3.nd.edu/~rwilliam/xsoc73994/index.html

For a brief overview of modeling nonlinear relations, see

https://www3.nd.edu/~rwilliam/stats2/l61.pdf

As Phil says, it is not a good idea to ask people to do your homework, but asking for things to read (or maybe help with some substep where you've done a lot of work already but are now stuck) can be ok.

[Weiwen Ng]

I don't mean to pile on here, but the command mfx has long been superseded by the margins command. In any case, I don't think margins will get you what you want for that question, either, because you transformed the original variables. Think about what logging a variable does to the variable. For example, try

scatter y ln_y