I was trying to find out the effect of neighbourhood characteristics on the losses sustained in a flood disaster in terms of income, farm produce, properties, lives, farmlands and displaced persons . Using ANOVA, I found a significant difference in household losses across the five neighbourhoods. I subsequently ran a Tukeys' post hoc test to account for these variations. However, the size of the result table is bigger than what can conveniently be accommodated within my text. Is there any way I can reduce the size, or summarize the contents while still keeping the essential parameter that would help in my explanation?
Perhaps, it is worth to show only the significant pairwise comparisons.
Kinds,
The usual approach is to use a compact letter display (cld) or a similar graphical technique.
The procedure to make a cld goes through each comparison, and assigns a letter to each group --- in your case group x treatment interaction, I imagine ---. Groups with the same letter are not significantly different at some alpha value (say, 0.05).
So, to take the example from http://rcompanion.org/handbook/I_08.html , and look at the variable Supplement, the six pairwise comparisons get reduced to:
Supplement Mean .group
C 1146.667 a
A 1153.333 a
D 1263.333 ab
B 1313.333 b
This can be represented graphically, eg:
http://rcompanion.org/handbook/images/image130.png
Although you may want to use an interaction plot style, eg., but with the mean separation letters added:
http://rcompanion.org/handbook/images/image131.png
You have to be careful with the interpretation. For example, in the above table, Supplement D is not different than A and also not different than B, even though A is different than B.
* * *
SAS used to produce a compact letter display by adding the LINES option to the MEANS statement, I think. It's common output from LSMEANS, I think.
R can produce them with, eg. multcompLetters in multcompView or cldList in rcompanion. Search for “cld” or "compact letter display" here for examples: http://rcompanion.org/handbook/
The size of the result table will definitely be large; however, you only need to pick out the comparisons that are of interest to your study. Normally,no one expects you to display the entire text that could span several pages.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA