Foo - Any statisticians in the house?

Or anyone else versed in statistics and spss? If there are any, HALP!!!!!!!!!!!!

I'm trying to analyze some data and can't consult with my adviser until next Monday, but I want to check out some results now so I can include it in a draft that I'm going to go over in a scientific writing class. I'll add more details about the experiment if someone actually owns up to being a statistician.

You should post a little more detail about what you need.

58% of all statistics are made up on the spot.

You should post a little more detail about what you need.

I will assume you are good at statistics...:p

I'm doing a study comparing the effects of three different types of primes (music, sentence, none) on the recall of words related or unrelated to the prime. Basically it's a 2x3 setup. All subjects go through each condition twice.

First problem: how do I combine the two instances of each condition? Adding? Average? Problem is that each time the trial is repeated for the condition the number of recalled words vary greatly (as a function of word length, familiarity, etc.). Or should I not combine them at all and just have them as separate instances?

Second problem: How the #*%^ do I input it into spss so that it makes sense? I've lost touch with spss since I've been using matlab and excel in our lab, but I have to use spss because my adviser for this project uses it it. This is what it currently looks like in excel:

http://lh4.ggpht.com/_SdF0uuk3iUc/ShNI-FOAulI/AAAAAAAACKs/2lHwngiWEGg/s800/data.JPG

58% of all statistics are made up on the spot.

Nu uh, it's 73%. Get your stats straight :lol:

Haha, I'm no statistician. I just suggested you post a little more information because in my experience, and I have plenty of it when it comes to asking obscure questions, is that more information fetches better quality replies. I took my first stats course last semester :p

Haha, I'm no statistician. I just suggested you post a little more information because in my experience, and I have plenty of it when it comes to asking obscure questions, is that more information fetches better quality replies. I took my first stats course last semester :p

*facepalm*

Then again, I'm on my third stats course and fourth lab class, but I still can't figure stuff out because this is a totally new paradigm.

If you're doing an ANOVA, you should just input each observation seperately. If you're just trying to compare numbers, averaging is sensible. Psychology experiments always have tons of variation, which is why you really need more than two observation per condition.

For the second part, I've never really used SPSS, so I could be wrong here, but typically the right thing to do is to add a line for each observation with a columns for conditions and result. So, your data file would look something like this:

ID | Prime | Related | Words
----------------------------------------------
1 | 'Music1' | 'Yes' | 4
...

If SPSS doesn't let you use strings for variables, assign each condition a number (Related = 1, Not Related = 0) and set the variable type to factor/categorical.

Hopefully that should help.

Normally, for each condition, I'd have a binary answer set of 1 or 0 for yes/no. Tally the results and divide by the number of subjects to give a decimal correlate.

First problem: how do I combine the two instances of each condition? Adding? Average? Problem is that each time the trial is repeated for the condition the number of recalled words vary greatly (as a function of word length, familiarity, etc.). Or should I not combine them at all and just have them as separate instances?

You shouldn't combine them, your results are telling you you have as much or more variation within subjects as between treatments (conditions) if I'm following you correctly.

When each subject gets each treatment twice, do they get the same words to recall each time? If they don't, does each subject get the same words per treatment? (is the 1st time through the same words for each subject, 2nd time through, different words from 1st time but same for each subject).

You've got repeated measures here.

And, do you have an IRB to conduct this experiment?

You've got repeated measures here.

And, do you have an IRB to conduct this experiment?

Hmmm...yes, but I am having trouble arranging the data in spss so it makes sense when I use the repeated measures. I'll figure it out In any case, prime would be the within-subjects variable and relatedness would be the between, right?

And yes I have irb approval and all that for my "heinous" experiment :p


You shouldn't combine them, your results are telling you you have as much or more variation within subjects as between treatments (conditions) if I'm following you correctly.

When each subject gets each treatment twice, do they get the same words to recall each time? If they don't, does each subject get the same words per treatment? (is the 1st time through the same words for each subject, 2nd time through, different words from 1st time but same for each subject).

Nope, individual subjects don't get the same words in each trial, nor do different subjects get the same words per condition.

Hmmm...maybe I should be looking at differences between conditions instead?

model it I say

I don't think you would combine the values of the two instances. (Ideally you will want more than two instances in each condition, actually, because two does not allow analysis of variance, and I think that ANOVA is the only method you can use in a 2x3 set...)

Basically what you need your stats to tell you is whether there is a significant, overall difference in the number of words recalled under one condition compared to the others. To do that you first have to analyze the amount of variation *within* each condition, which is where the ANOVA comes in. I think you'd do it one-way here. Your data set looks a bit like one of the ones we used in my last Stats class and that's what we did.

I remember being really confused by the data table in SPSS - something I thought should go across went down, or vice versa. I think IanHelgesen's suggestion is a good one - set up columns and assign strings for the variables (it does let you do that), with the number of recalled words being your only numerical value. I do remember that the help file in the SPSS program was surprisingly sane and useful.

Hmmm...yes, but I am having trouble arranging the data in spss so it makes sense when I use the repeated measures. I'll figure it out In any case, prime would be the within-subjects variable and relatedness would be the between, right?


This depends on the experimental setup. From what I can tell, all subjects are in all conditions, which means that both variables are within-subjects. Basically, if you have observations of an individual across multiple conditions, you can use that information to remove individual variation, leaving you with a better estimate of the effect of condition.

You'll also need to give each subject an id number, so that you can group observations by subject. Part of the problem here mayalso be that the reapeated measures procedures are probably not what you want. Repeated measures are generally for when you need to incorporate time into your model. For a simple within-subjects (hopefully counterbalanced) design, you should be able to select the appropriate options from the ANOVA dialog (assuming you have enough data for an ANOVA).

(Hopefully that's all correct. My applied stats have gotten a bit rusty these days, and I'm too lazy to go pull out my textbooks.)

It's true: 100% of people who eat carrots die.

Ok...gonna start wading through data now. Thanks for your suggestions, guys! :D

Normally, for each condition, I'd have a binary answer set of 1 or 0 for yes/no. Tally the results and divide by the number of subjects to give a decimal correlate.

+1 or 0 or something....but this is how I would approach this.