Hi Rich!
First, thank you for your answer.
Am Tue, 19 Feb 2008 18:30:21 -0500 schrieb Richard Ulrich:
> On Tue, 19 Feb 2008 16:52:21 +0000 (UTC), Martin Kaffanke
> <martin.kaffanke@[EMAIL PROTECTED]
> wrote:
>
>> Hi there!
>>
>> I have 2 data sets from the same 20 items. One is made by myself, I
>> have 24 persons tested on the computer with this items.
>>
>> And there is another work where I have the data from, which used the
>> same items but on paper/pencil, where 47 persons were tested.
>
> Okay. You have a test administered in two ways, paper/pencil and on
> computer. You want to compare something.
>>
>> There are also correct answers to the 20 Items.
>
> You want to compare whether one approach gets better answers (assuming
> the groups were equivalent).
Yes, the groups weren't relly equivalent, but that doesn't matter,
because I can say group A has better results than group B because of the
computer and the other differences. I know about this lack.
>> Oh, and I have to say
>> this are really small probabilities like .00001 and the scale for the
>> answers was logarithmic.
>
> "The scale of the answers was logarithmic" sounds like the answers were
> quantitative. Rather than the usual, Right /Wrong, they were graded by
> how far off they were, multiplicatively.
Yes, I do a "item A"/"correct" transformation which tells me that the
value of 1 would be correct and the difference from 1 is the discrepancy
of the correct value.
>
> "really small probabilities like .00001" comes out of the blue, and
> does not make any sense.
Ok. So lets say the scale was like:
- 1:1
- 1:10
- 1:100
till
- 1:1000000
this was logarithmic, so I think i should also do some logarithmic
transformation. The values I got are now 1/100 if the user choosed
1:100, it was also possible to do select any value between the given
items. If you'd like to have a screenshot or something, just tell me.
>> Now I'd like to see if my computer test provides better results then
>> the other one.
>>
>> How can I do that?
>>
>> I'm using R as statistic software. How can I compare this?
>
> If you have a summary "Grade" for each person, as if these were
> schoolroom tests graded from 0-100, you do a t-test on 24 versus 47
> subjects.
> If you don't have a grade, you have to create one.
What do you mean by grade?
> Does that say enough?
Yes, the t-test tells me a good start, but does it fit to logarithmic
data? As I can see the linear mean of
..1 and .001 is 0.0505
But when I look at the scale I had, the mean of the linear value (by
measurements of a ruler) would be 0.01 so I think I have to do any
logarithmic transformation of the values to get good results.
Thanks for your help,
Martin
|
