Aha! I think Brian has got it.
Sometimes we (and I mean me) obsess about doing the "right"
significance test. When maybe we just need to show what happened. A
useful thing to remember is that Piaget, Milgram and Skinner never did
a significance test.
Jeremy
On 9 July 2011 22:09, Brian K. Saxby <[log in to unmask]> wrote:
> Hi Laura,
>
> It looked to me like it should be simple, although I didn't know the answer,
> so I guess we can all take some reassurance in that it's definitely not
> simple if Jeremy says it's tricky!
>
> How crucial is it that you have some statistical analysis behind what you
> want to say? Could you get away with simple descriptions of the data? One
> way to present it visually is a 100% stacked bar graph (could even use Excel
> for this) - each bar represents a food, and the graph will show the
> proportion it was rated as hi vs low vs don't know, and you can compare
> across foods. If you stack the data so they are hi, don't know, low, and
> sort your data in order of descending hi, then ascending low values, you
> should be able to see clearly the hi carbon footprint foods at one end and
> the low carbon at the other. The middle will be something of a risotto
> though.
>
>
> On 09/Jul/2011 13:13, Jeremy Miles wrote:
>>
>> This is a very tricky question. (If you don't mind my saying, it's a
>> good example of thinking about your analysis before you collect the
>> data.)
>>
>> I guess a Friedman's non-parametric repeated measures ANOVA is the way
>> to go, but I don't know if that will work, I've never seen it done
>> with more than three or four variables. Then your post hoc tests are
>> going to be tricky (and hard work, there's a function in R that will
>> do them automatically, but if you get a significant result you're
>> going to need 70*69/2 = 2415 (if I've counted right) tests. Which is
>> going to get boring.
>>
>> A better way would be to fit a multilevel ordinal logistic model, with
>> using a reference coding scheme, which will compare each food to the
>> average. I'm not sure such a thing exists (it can't be done with
>> SPSS, I'm pretty sure it can't be done with Stata, SAS or R. You
>> might manage in HLM or MlwiN [Mlwin is free if you're a UK academic]
>> but it's not easy).
>>
>> A complex samples approach in SPSS might do the trick (google Huber
>> White SPSS to find out how to do it), if it will do ordinal logistic
>> regression with complex samples (I'm not sure if it will, I've never
>> tried), or you could also use generalized estimating equations in SPSS
>> (Or SAS, Stata or R). [Again, I'm not positive that this actually
>> exists. The usual solution if it doesn't is to build up dichotomous
>> logistic regressions, but that won't give the post hoc tests you want.
>> Or if it will, I don't know how to do it).
>>
>> None of these are very easy. It would help if there were someone at
>> your institution who knew how to do one of them. Where are you?
>>
>> Jeremy
>>
>>
>>
>> On 8 July 2011 15:13, Laura Sale<[log in to unmask]> wrote:
>>>
>>> Does anyone know the best way to analyse this data??
>>>
>>>
>>>
>>> We had 42 participants who were given 70 different food items and they
>>> had
>>> to assign each item to one of three categories:
>>>
>>> 1. Low Carbon Footprint
>>> 2. High Carbon Footprint
>>> 3. Don't Know
>>>
>>> For each participant we have recorded how they categorised each item
>>> (low/high/don’t know) and so for each food item we will have frequency
>>> data
>>> for how often that particular item was categorised as either
>>> low/high/don’t
>>> know by participants. What would be the best way to analyse this data to
>>> find out if certain food items are more likely to be classed as having a
>>> high carbon footprint and some are more likely to be classed as having a
>>> low
>>> carbon footprint?
>>>
>
|