Hi.
Does the level of measurement on the scale make a difference to an
indication for a parametric test too? My understanding is that the
outcome measure should be an interval or ratio level of measurement
and that sometimes, a likert scale can also be used (open to debate
though), depending on the operational definitions or way that the
scale is constructed...is that correct? I dont want to raise
questions on an incorrect basis or lead you up the wrong path.
A question about the ranks in a non-parametric test: the SPSS output
tells you which ranks are negative, positive and tied. Can you look at
the descriptive stats, see which way the change occurs and then
interpret the test results, with that in mind?
Best regards to all. Claire.
Quoting Jeremy Miles <[log in to unmask]>:
> Hi David,
>
> If you're planning to do a repeated measures t-test (and it sounds
> like you are) then it's not the variables that need to be normally
> distributed, it's the differences (or equivalently, the residuals).
> Even if the variables are highly non-normal, the differences might
> still be normal. It's also possible that you could do a
> transformation to normalize your data.
>
> However, if you have a large (say more than 100) sample size, then you
> don't need to worry about normality.
>
> How does using a non-parametric test limit the number of cases you can
> use? I've never heard of that, and would be very averse to analysing
> less than all of the data.
>
> Non-parametric test results are much hard to interpret than parametric
> , because you don't have a meaningful interpretation, except that
> there are differences in ranks - with a t-test you can say that the
> mean went down X points. I do statistical analysis all day long on
> this kind of data (it's my job), and I rarely use a non-parametric
> test (I can't remember the last time I did).
>
> Jeremy
>
>
>
> 2008/12/23 David Eley <[log in to unmask]>:
>> Hi everyone,
>> I am analysing some outcome measures for a Chronic pain and fatigue
>> management service and having some trouble deciding whether I can use a
>> t-test or not.
>>
>> Very simple with ask patients to fill in a variety of outcome measure
>> questionnaires pre treatment group and post treatment group. I just want to
>> see if our treatment has had a significant effect and helped them.
>> I was going to use a T-test, however our data does not meet all the
>> parametric assumptions; specifically the one for having normally distributed
>> data. So have been advised to use a non-parametric equivalent. However this
>> limits the amount of data we can use from the database.
>>
>> As this analysis is not for research purposes can I still use the t-test so
>> that we can include more data, even though the data does not meet all the
>> parametric assumptions?
>>
>> My reasoning being that we try and make our samples meet the parametric
>> assumption because we wish to generalise our results to the general
>> population where we believe these assumption exist. However, we are not
>> generalising the general population as is this still essential for the data
>> to be normally distributed. Or if we are generalising then it would be to
>> the population who suffer chronic pain and fatigue, who would not be
>> normally distributed on these outcome measures anyway.
>>
>> The one flaw that I can potentially foresee is that perhaps there is a good
>> mathematical reason for why the t-test result would be invalid? i.e. the
>> actual maths formulae needs the data to be normally distributed for it to
>> run reliably?
>> If anyone has any thoughts (though preferable answers!) I would be very
>> grateful.
>>
>> Merry Christmas
>> Dave
>>
>> David Eley
>> Assistant Clinical Psychologist
>> Community Neurological Rehabilitation Team and
>> Chronic Pain and Fatigue Service
>> Rayners Hedge
>> Croft Road
>> Aylesbury
>> Buckinghamshire
>> HP21 8HZ
>> Email: [log in to unmask]
>>
>>
>> ________________________________
>> Take your friends with you with Mobile Messenger. Click Here!
>
>
>
> --
> Jeremy Miles
> Learning statistics blog: www.jeremymiles.co.uk/learningstats
> Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
>
_____________________________________________________________________
Claire Louise Russell
|