svy test interpretation

[Claudia Contreras]

Hello! I'm working with svy data and I'm having difficulties with test interpretation.

I created two globals:

global one "variable_a variable_b"

global two "variable_c variable_d variable_e"

Then, I regress my dependent variable with other variables and the globals as covariates:

svy: reg variable_x variable_y variable_z $one $two

After that, I test de covariates.

test variable_y
test variable_z
test $one
test $two

How would I must interpreted the results of F-statistics?

Thanks!

[Claudia Contreras]

The p-value indicate if the variable or the global is significant?

[Steve Samuels]

Welcome to Statalist, Claudia

Generally the answer is "yes"; the manual entry for test describes all the options. But with svy data, the tests will be in error if you used a finite population correction (fpc),

FAQ 12 is essential reading for new members. If you haven't read it, please do so now. In future posts, show us all the relevant commands and their results, pasted between CODE delimiters, described in the FAQ. In a question about survey data, be sure to show your svyset statement.

[Claudia Contreras]

Thanks, Steve!

I would try to improve my post.

My situation is this:

svyset idschool [w=wgt], strata (idstrat)

global one "variable_a variable b"
global two "variable_c variable_d variable_e"

generate sp1= 1
replace sp1=. if variable_x==.
replace sp1=. if variable_y==.
replace sp1=. if variable_z==.
replace sp1=. if variable_a==.
replace sp1=. if variable_b==.
replace sp1=. if variable_c==.
replace sp1=. if variable_d==.
replace sp1=. if variable_e==.

svy, subpop (sp1): reg variable_x variable_y variable_z $one $two

test variable_y
test variable_z
test $one
test $two

For test $one, for example, I obtained:

( 1) variable_a = 0
( 2) variable_b = 0

F( 2, 121) = 7.41
Prob > F = 0.0009

I understand that this mean that the variables "a"and "b" are statistically significant for explaining variable x.

Is that correct? How should I must interpretated the results?

[Steve Samuels]

You are correct:. I'm curious about the tests. What is it that the variables var_a and var_b have in common that leads you to test them jointly? The same question applies to the variables in $global2. Note that there is no real need to test individual variables as the p-value already appears in the regression output. In fact, it can be dangerous unless you care about each one: as the number of hypothesis tests goes up, the chance of a false positive increases (the multiple test problem).