% of Increase vs. P-Value

#1
Hey guys,

In my research, I have found a significant difference (p<0.05) in a behavioral tendency after a psychological intervention. By looking at the scores, it becomes clear that this difference is an increase. I have 2 questions:

1) Can I give the percentage of participants who showed an increase? I feel like the p-value gives more credibility than a percentage which does not really prove anything. Am I wrong?


2) If it is relevant, how can I do it in SPSS?



Thanks a lot in advance :)
 

obh

Active Member
#2
Hey guys,

In my research, I have found a significant difference (p<0.05) in a behavioral tendency after a psychological intervention. By looking at the scores, it becomes clear that this difference is an increase. I have 2 questions:

1) Can I give the percentage of participants who showed an increase? I feel like the p-value gives more credibility than a percentage which does not really prove anything. Am I wrong?


2) If it is relevant, how can I do it in SPSS?


Thanks a lot in advance :)
I don't understand exactly what results you got. What test did you run? Is it comparing to proportions? %increase before and %increase after?

Any way,
P-value is just one parameter... And doesn't come instead of presenting the research results..you can also present the effect size
 
#3
I don't understand exactly what results you got. What test did you run? Is it comparing to proportions? %increase before and %increase after?

Any way,
P-value is just one parameter... And doesn't come instead of presenting the research results..you can also present the effect size

Thanks for your response:)

I did a paired samples t-test based on pre- and post-scores on a survey.

I am a bit unsure if I interpret the meaning of percentages vs. p-value correctly. When I have a significant p-value, then it means that the increase in scores post-intervention are likely not due to chance. But if I give a percentage now of the number of participants whose score increased, can I reliably say that these scores are not random/due to chance too? Is giving a percentage a good idea?
 

obh

Active Member
#4
Pair t test compares the averages of before and after. So you should first show these averages. (and /or the average improvement in score) generally this should be enough. You can also show confidence interval for the proportion of people with increased value.

" percentages, can I reliably say that these scores are not random/due to chance too..? "

If you assume the the differences (after - before) distribute t (the sample the population z) with average grater than zero, than you can calculate p(x>0) and it will be more than 50%.

http://www.statskingdom.com/41_proportion_confidence_interval.html
 
Last edited:
#5
Pair t test compares the averages of before and after. So you should first show these averages. (and /or the average improvement in score) generally this should be enough. You can also show confidence interval for the proportion of people with increased value.

" percentages, can I reliably say that these scores are not random/due to chance too..? "

If you assume the the differences (after - before) distribute t (the sample the population z) with average grater than zero, than you can calculate p(x>0) and it will be more than 50%.

http://www.statskingdom.com/41_proportion_confidence_interval.html
Okay I understand, so calculating that will give me more certainty what the true difference between the scores is. Thank you for your help!
 

obh

Active Member
#6
I assume you talk about the percentage of improved subjects . Theoretically, based on normality assumption, if the difference between mean is significance, the percentage of improved subjects should be greater than 50% and can be calculated based on the t distribution, but I would do confidence interval.