2 more questions on the 2nd class handout
1. In the 2nd class handout, page 2, the first slide, the Levene's test came out significant so I looked to the bottom column to find the t-value. In reporting this t-value though, the degree of freedom is not a whole number. Is there any rule for reporting df values that has decimal points?
- Can you reiterate how to report the t-values again?
2. In the 2nd handout, page 3, the third slide, Dr. Clemente did a post-hoc test for the little league groups and used a Dunnett T3. However, the data did not violate assumptions of homogeneity and the N = 191, clearly bigger than N<75. Can you explain why she used the Dunnett post-hoc instead of the others in this case?
Thanks!
- Can you reiterate how to report the t-values again?
2. In the 2nd handout, page 3, the third slide, Dr. Clemente did a post-hoc test for the little league groups and used a Dunnett T3. However, the data did not violate assumptions of homogeneity and the N = 191, clearly bigger than N<75. Can you explain why she used the Dunnett post-hoc instead of the others in this case?
Thanks!
posted by Narae @ 4:38 PM
2 Comments:
At 6:19 PM,
Mari Clements said…
The information about how to properly report results of t tests is actually available to you in the APA Publication Manual. (The Publication Manual should be your first line of inquiry for all formatting questions.)
Having said that, the general form is:
t(df) = x.xx, p = .xx
where
1. t and p are both in italics
2. df is the actual degrees of freedom, not letters
3. the t is reported to two decimals
4. if the t value is less than 1 but more than -1, a zero is in the one's place
5. if the t value is less than zero, a negative sign is included
6. the p is reported to 2 or 3 decimals
7. the spacing is exactly as above
8. some index of effect size (d or CI) is also included
As for fractional dfs, there's no real consensus. Some people round (typically down), some report the df to one decimal, and some report it to two decimals.
As a side note, t tests are not a central focus of this class: they are only addressed in relation to ANOVA.
At 6:52 PM,
Mari Clements said…
2. In the 2nd handout, page 3, the third slide, Dr. Clements did a post-hoc test for the little league groups and used a Dunnett T3. However, the data did not violate assumptions of homogeneity and the N = 191, clearly bigger than N<75. Can you explain why she used the Dunnett post-hoc instead of the others in this case?
---
Actually, that slide (which we did not get to) is a comparison of three post-hoc multiple comparison tests: LSD, Dunnett's T3, and Scheffe...none of which were actually appropriate.
It is true that Fmax does not exceed 10 (it is 6.43), and N is above 75. Thus, in this case, the T3 test was more conservative than necessary (i.e., provided more correction than perhaps was needed).
Neither Tukey nor Scheffe (as offered by SPSS) are good choices because they both assume equal cell sizes (exactly equal, not approximately equal).
LSD is never a good choice, because it is equivalent to uncorrected t tests (and thus inflates alpha).
The better choice would likely have been a Bonferroni, because (a) assumes homogeneity of variance and (b) does not assume equal cell sizes.
The questions about multiple comparison procedures on the test are more general and/or more scaffolded than this. You will not be asked to decide among specific post-hocs, although you might be asked whether a post-hoc should be conducted and/or whether you should use one that assumes equal variance or not and/or how to read a post-hoc output from SPSS.
Post a Comment
<< Home