1
$\begingroup$

Me again asking the silly questions - this might belong here as I've been told it may belong in a software section somewhere..? Which I have still been unable to find.

Here is a screenshot of my data: enter image description here

Thanks to an awesome answer to my last question I was able to calculate the variable I wanted - which is labeled YrmthASpsec (in screenshot). This is the percentage of time spent in each feeding event on a monthly basis (year>month) by age-sex category.

No what I would like to do is compare this via ANOVA - however I am having much troubles.

I would like to compare the mean percentage on a monthly basis for each age-sex category by plant part. So... compare mean % of time of fruit consumption (number 1) between age-sex classes for April 2002 for example.

I have fiddled with this for a while but with no success.

Need more info? Just ask :)

EDIT1

I am seeking to determine whether there is a significant difference in dietary composition between age-sex categories. Furthermore I am also looking at nutritional composition (macro) so protein, lip and sugar across age-sex classes.

So looking for the best way to do that using the data presented above. Most sources I have consulted use time as a proportion as it indicates importance/investment in a food item.

$\endgroup$
7
  • $\begingroup$ If you can frame your question clearly as a statistical one (rather than simply 'what commands do I need to do X'?) then it definitely belongs here. If it's effectively a 'what commands' question, suitable places for help on using stats packages are listed here; the specifc-to-SPSS answer is then here $\endgroup$
    – Glen_b
    Commented Feb 24, 2015 at 23:49
  • $\begingroup$ As a general comment on your problem, with compositional data, including "percentage of time spent in activity" type DVs, it's very common to have skewness (since those percentages may often be small) and heteroskedasticity (since the expected percentage tends to change in different conditions, and the spread in a bounded variable will generally be related to the mean). There are more appropriate models for that kind of data (beta regression is a common choice, for example). ...(ctd) $\endgroup$
    – Glen_b
    Commented Feb 24, 2015 at 23:55
  • $\begingroup$ (ctd) ... If it makes sense for your application, a transformation may be useful (e.g. sometimes log of proprotion, or log of odds of being engaged in the activity may be reasonable to consider and may be less skew and less heteroskedastic), but I wouldn't necessarily go that way. If the percentages are not typically all that small you may be okay with ANOVA, though it might be necessary to consider a Welch-type adjustment for heteroskedasticity. $\endgroup$
    – Glen_b
    Commented Feb 24, 2015 at 23:57
  • $\begingroup$ please see edit. I have been told so many different things I am lost. I understand stats is a huge field and there is still much debate but I have one person (not on here, just in general) telling me I can't do it because of skew and another saying because it is all skewed in a certain way that it should be fine. :( I find it all so confusing. $\endgroup$ Commented Feb 25, 2015 at 1:42
  • $\begingroup$ Wait ... so someone said to you that if it's all similarly skewed ANOVA should be fine? That's not generally the case. Did they provide any kind of support for the claim? $\endgroup$
    – Glen_b
    Commented Feb 25, 2015 at 1:50

1 Answer 1

0
$\begingroup$

Beta regression is available in SPSS Statistics via the STATS PROPOR REGR extension command. That can be installed from the Utilities menu with V22 or later or downloaded from the SPSS Community website (www.ibm.com/developerworks/spssdevcentral) in the Extension Commands Collection. See that item for details.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.