test and testnl - can they be used in that case? (comparison on specific predicted points)

[Luc Meunier]

I have been struggling on a problem from the past few days, and I hoped you could help. I have estimated an OLS interval regression model (intreg) of the form:

Y=β0+β1.x+β2.ln(x)+...

β1 and β2 are significant and have opposite signs.

β1=0.0053

β2=−0.2464

When I plot their joint effect, I obtain this plot (x is bounded between 1 and a 100).



I would like to show with a nice statistical test that an increase in x leads to a significant decrease in Y from 1 to 46 and that it has no significant overall impact afterward (from 46 to 100).

I was thinking I could use the command test (or testnl, even though not necessary in that case). For instance, to test the underlined part above, I would test

H0: β1.46+β2.ln(46) = β1.100+β2.ln(100)

And type in Stata:
test (46*x+LNx*3.8286=100*x+LNx*4.6051)

Is it a correct use of the command test?

[Phil Bromiley]

You'll increase your chances of a useful answer by following the faq on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

What you're doing works with the parameters but not really with the predictions. I'd look at margins with the contrast option.