Home › Forums › Default Forum › Non-Parametic ANCOVA
- This topic has 8 replies, 4 voices, and was last updated 8 years, 12 months ago by
Jeremy Miles.
-
AuthorPosts
-
24th January 2012 at 2:54 am #2804
Donna M
ParticipantHi Andy
Does the bootstrapping function on SPSS include the ability to do an ANCOVA type analysis on non-parametric data?
Or are there any other ways to go about this? I looked at the R plug in and it seemed really complex so I was looking for a more straight forward approach.
Thanks,
Donna.
24th January 2012 at 4:32 am #2812Jeremy Miles
ParticipantData aren’t really non-parametric, tests are.
ANCOVA is inherently parametric. Bootstrapping is parametric (because you’re estimating parameters).
But to answer your question, ues, you can bootstrap an ancova which means you can relax the normality assumptions.
24th January 2012 at 5:08 am #2811Donna M
ParticipantThanks Jeremy for your response, that’s great.
I had heard that the bootstrapping functionality in SPSS is not a standard part of the programme but that it is sold as a separate package – do you know if this is the case?
24th January 2012 at 5:14 am #2810Anonymous
Inactivenow… why is it that you’d like to try a non-parametric analogue of ANCOVA? the normality assumption to perform inference in the case of any linear model is on the residuals and not on the variables themselves… could you perhaps tell us a little bit more about the data you are working with?
24th January 2012 at 5:32 am #2809Donna M
ParticipantMy data is not normally distributed and I’d like to be able compare results for 2 groups on a measure whilst controlling for another variable.
24th January 2012 at 6:25 am #2808Anonymous
Inactiveonce again, the assumption of normality for infernce is not on the variables themselves but on the residuals after you’ve performed your analysis… have you tested those to see whether they look normal or not?
24th January 2012 at 6:40 am #2807Donna M
ParticipantThe tests of normality I have undertaken is the Kolmogorov-Smirnov test on the distribution of the results for the variable as a whole, rather than the residual.
24th January 2012 at 7:04 am #2806Anonymous
Inactivewell, test the residuals then and we’ll take it from there if they violate the normality assumption…
29th January 2012 at 8:06 pm #2805Professor Andy Field
MemberAs oscar says, test the residuals (remember with tests like K-S if you have a large sample then they’re pointless because even small deviations will be deemed significant, and with small samples the tests have low power, so really graphs and values of skew/kurtosis are probably most useful).
If you have a problem, then from what I recall the bootstrapping module (it was a separate module, but I think under IBM it is integrated into the core package) only bootraps SEs on the mean for things like ANCOVA.
You have more options in R to be honest, but if you want to use R I wouldn’t bother with the SPSS R plugin because you may as well just use R directly.
good luck,
a.
-
AuthorPosts
- The forum ‘Default Forum’ is closed to new topics and replies.