Interpreting results from mfx

[Zuhumnan Dapel]

Please help with idea on interpreting results from marginal fixed effects. I run a simple regression. Dependent variable is 't' ( natural log of years) and independent variable is 'g' (growth rate in percentage). After 'margins, at(g=80), command, I got the value of 0.4354895. Can someone please tell me the meaning? Thanks

Commands
reg lnt g, robust


. margins, at(g=80)

Output

Adjusted predictions Number of obs = 2549
Model VCE : Robust

Expression : Linear prediction, predict()
at : g = 80


| Delta-method
| Margin Std. Err. t P>|t| [95% Conf. Interval]

_cons | .4354895 .0245527 17.74 0.000 .3873442 .4836347

[Carlo Lazzaro]

Dear Zuhumnan,
thanks for re-registering as per FAQ.
Please find replies to the thread you started at http://www.statalist.org/forums/foru...sults-from-mfx.

Kind regards,
Carlo

[Zuhumnan Dapel]

Thanks Carlo. I still need some help on what I posted.

[Maarten Buis]

You need to give more details than that. What did you not understand about these answers? Especially, the answer by Rich, including the links, looks like a pretty comprehensive answer to me.

[Zuhumnan Dapel]

Thank you all. @Maarten What specific details are required apart from the commands and the output I provided?

[Nick Cox]

Zuhumnan: There are issues at many different levels here.

1. You don't provide a reproducible example. We can't see the dataset you are using therefore. Nor, in various questions that you ask, can we always see exactly what commands you are using.

2. In your example, you seem to be confusing dependent and independent variables, both in the terminology you use and in how you are using margins. Thus although you may have an unusual reason for inverting the analysis, in most research time is regarded as an independent variable and something else which depends on time, in your question growth rate, is thus a dependent variable (better terminology: response or outcome variable).

3. Even correcting terminology, there is still a puzzle of why you think it makes sense to

3.1 work with natural log of years (that makes little or no sense because for the purposes you presumably have the calendar origin is arbitrary)

3.2 ask for margins "at" a value of the response variable (the help makes it clear that the typical application is "at" a value of a covariate)

4. More generally, asking for help in interpretation, or on what something means, is usually posing a difficulty because it is not clear what you don't understand or what kind of answer you are seeking. Sometimes students or junior researchers want the substantive (e.g. scientific or social scientific) interpretation, which is often difficult to provide and in any case much of what they should be producing for themselves.

It's crucial to understand that unless people answering on Statalist are StataCorp employees, we are all volunteers and busy with our own work, and so unlikely to answer unless we can answer easily and quickly. If someone is very unclear, or appears very confused, or appears to be asking the same questions repeatedly, people are less inclined to wade in.

[Richard Williams]

As Nick points out, substantively, the regression doesn't seem to make sense. But regardless of whether it makes sense or not, mechanically, it is not hard to figure out what the margins command is doing. You didn't show the output from the regression command, but assuming you have described it correctly it should be the case that

80 * coefficient for g + constant term = .4354895.

So as Nick suggests, I would suggest you start by estimating a model that makes sense. Then if you still want to use margins, read through the handouts I suggested earlier, and then if you still have questions be specific about what the problem is, e.g. "I am unclear about what the difference between marginal effects at the means and average marginal effects is." As it stands, we are not clear about what it is you are unclear about.

[Zuhumnan Dapel]

Dear Nick and Rickard thank you much for the replies. First, with respect to the dependent variable, I'm not referring to time in the literal sense of as you have understood. I 'm referring to average exit time from poverty, which I computed from a given household survey dataset. The time depends of the rate of growth in mean consumption of households. Second, in the initial post, I have posted the results. I'm so sorry that the command and the results I posted were not visible to you.

[Richard Williams]

Thanks for clarifying what time means. This makes me wonder if you should be doing a survival model instead. Beyond that, I am not clear what the remaining question is. I offered an interpretation a few posts earlier and in the linked thread I offered some readings. So, if you can clarify what remains unclear we can try to help you more.