Two sample unpaired ttest with mean difference specified

[matt Mosh]

I need to run a t-test on a two sample group that is unpaired with unequal variance for a specific mean difference.

The data sample is unpaired and my research hypothesis is:

Upstream- Downstream “at least” (one sided) ≥ 0.5ppm

I found a note that STATA cannot do this:

Unlike the SAS TTEST procedure, the .ttest command cannot specify the mean difference D₀ other than zero.

http://citeseerx.ist.psu.edu/viewdoc...=rep1&type=pdf

If the data were paired I could create a new variable with the difference and run a ttest on that for the specified ppm.



The "R" code is:

t.test(Upstream,Downstream,alternative="greater",m u=0.5,conf.level=0.99)

Thanks,

Just a Noobie, Matt

[Clyde Schechter]

So you can emulate this with a regression and the -lincom- commands.

Let's say that the variable that distinguishes "upstream" from "downstream" is called location, coded 1 for upstream and 0 for downstream, and your outcome is called outcome.

regress outcome i.location
lincom _b[1.location]-_b[0.location] - 0.5

[Richard Williams]

Welcome to Statalist.

I think you could also just add or subtract a constant from one of the groups, e.g. subtract .5 from the group that you think is greater by .5.

Also with Clyde’s, I think 0.location = 0. So, a simpler statement might be

test 1.location = .5 (or else -.5)