|
From: | Alan Mead |
Subject: | Re: pspp-dev Digest, Vol 145, Issue 1 |
Date: | Mon, 4 Jan 2016 14:52:59 -0600 |
User-agent: | Mozilla/5.0 (Windows NT 6.1; WOW64; rv:38.0) Gecko/20100101 Thunderbird/38.5.0 |
This is a slightly longer answer. For those who are not aware, the
output for a two independent-sample t-test contains three
hypothesis tests. First, there is Levene's test for homogeneity of
variances and this is the "F" that Elias refers to. To the right
there are TWO rows: a t-test performed assuming equal variances and
a second row showing the same test conducted NOT assuming equal
variables. The PSPP output presents this exactly as SPSS does. My students are confused by this output; but it's probably largely due to them being novices. I think experienced users expect the output this way and are not confused. I think the current labeling is clear, but I would suggest a slightly different change with a header like (as Elias suggests) "Are equal variances assumed?" and then having values in the rows of "Yes (equal variances ARE assumed)" and "No (equal variances are NOT assumed)" But it might be even clearer (if less compact) if Levene's test were presented in an independent "box" and the t-tests were presented in their own "box". This is how SPSS presents Levene's and Box's tests for multivariate tests. Most clear of all would be to output something more textual that more closely matches how you would report the t-test... For example, if Levene's test were significant, then report the result NOT assuming equal variances and report it in a format more like what they should report: "Levene's test was significant (F = 10.1, p < 0.01; Var(1) = 10.2, Var(2) = 100.1) so the t-test with adjusted degrees of freedom was interpreted and found to be significant (t(37.48) = -5, p < 0.001; M(1) = -2.34, M(2) = 60.12)." Or if Levene's test were not significant, then just report the standard t-test: "The t-test was not significant (t(39) = -0.3, ns; M(1) = 52.34, M(2) = 60.12)." Maybe those should have effect sizes and confidence intervals as well. To do this, you would have to assume (or add a subcommand for the user to specify) an alpha level. Or, I guess, with just a little more coding you could report if the t-test were significant at the 0.05, 0.01 or 0.001 levels, as those are fairly standard. BTW, someone might want to know this: Levene's test is not considered very good because it is itself very sensitive to it's own assumptions. If you have equal samples sizes in the two groups, you may want to simply ignore Levene's test and the adjusted t-test and report the "standard" version (that assumed equal variance). -Alan On 1/4/2016 7:09 AM, Elias Estatistics
wrote:
Dear PSPP users, -- Alan D. Mead, Ph.D. President, Talent Algorithms Inc. science + technology = better workers +815.588.3846 (Office) +267.334.4143 (Mobile) http://www.alanmead.org Announcing the Journal of Computerized Adaptive Testing (JCAT), a peer-reviewed electronic journal designed to advance the science and practice of computerized adaptive testing: http://www.iacat.org/jcat |
[Prev in Thread] | Current Thread | [Next in Thread] |