I want to examine a factor structure of a behavioral questionnaire. The instrument consists of 8 clinical scales (total of 90 items), which, based on original (author's) factor analysis, form two indexes (or factors). Now, I want to conduct a factor analysis to confirm whether or not these 8 scales will form two factors in my sample of participants.
So, my question is: 1. What sample size do I need? Do I need to take care of all the items in the instrument (90 items= 450 participants or 8 items(scales), in which case 80 participants will be sufficient). My assumption is that clinical scales are correct and want to check whether these clinical scales form two factors (my goal is not to see whether 90 items will form 8 scales which further combine to form two factors).
a psychometric analysis, the normal sampling process is not useful, since people sometimes distort or selflessly respond to the questionnaire. Sometimes you have to remove the data relating to people who have responded poorly (the control items of the questionnaire, invalidate the participant), eg social desirability.
although there is no tacit agreement, to propose 7-8 participants per item, you have to analyze well. the variance of participant's response. if the questionnaire is well made, should be variance.
if you do a confirmatory factor analysis, should be increased to 20 subjects per item.
alternatives even as the Cronbach alpha omega McDonald, are more reliable with almost 1000 participants.
greeting
CJ
The sample size is determined by several factors, like the power of the study (90% or more), the size of difference that you want find in a measured outcome "delta" (also called clinical difference) and first type error "alfa" (usually set a 0.05%). First to calculate the sample size, the results reported in similar studies may provide a useful tool to take in to account an appropriate delta value of your scale.
First to combine all your observations in a factorial analysis, I suggest you to calculate the sample size in each clinical scale.
Have a nice day
As Paolo mentioned, sample size is determined by a number of factors. Confidence interval, confidence level and population size are the most important. These two site may be of some help:
http://www.surveysystem.com/sscalc.htm
and
http://www.raosoft.com/samplesize.html
a psychometric analysis, the normal sampling process is not useful, since people sometimes distort or selflessly respond to the questionnaire. Sometimes you have to remove the data relating to people who have responded poorly (the control items of the questionnaire, invalidate the participant), eg social desirability.
although there is no tacit agreement, to propose 7-8 participants per item, you have to analyze well. the variance of participant's response. if the questionnaire is well made, should be variance.
if you do a confirmatory factor analysis, should be increased to 20 subjects per item.
alternatives even as the Cronbach alpha omega McDonald, are more reliable with almost 1000 participants.
greeting
CJ
If the 8 scales are pre-aggregated in the analysis (i.e., you sum or average the 90 items into the 8 scales and factor analyze those), then a very safe number for a CFA would be 160 (8*20). Bentler advocated for 5-10 per parameter being estimated. I'm assuming that each of the 8 scales load onto one factor each (8 paths + 8 errors = 16 parameters). If you are going to factor analyze from the 90 items, then Carlos Jimenez-Gallardo is correct; you will need closer to 1000.
In reading the replies, I am getting the impression that the sample size guidelines being talked about are for a general purpose CFA. Haris, I take from your question that you are specifically testing whether two factors best fit the data which translates to a null hypothesis that some other number of factors best fit the data (or perhaps the other way around). This is a very non-standard test and I doubt that general purpose CFA sample size guidelines are going to work for it. Are you clear on what statistic you are going to use to test this hypothesis. That is a good place to start.
I would like to find out the factor structure of my 8 scales. The original inventory has two factors. I wonder how many factors will be extracted from my sample. Now, the question is should I take care of all the items (90) or just 8 scales. I believe that would lead my decision on sample size. I am aware from previous answers I need a larger sample for CFA but also was thinking about principal component analysis. The rule I read about in stat books is at least 5-10 participants per item. So, is it sufficient to have 80 participants for PCA or I need to have 450 participants? Basically I am not sure are my items 8 scales or all items in questionnaire (in which case I will not necesseraly have 8 scales).
I would stay away from PCA unless you have good measurement and a good reason to this the underlying factor structure is orthogonal. I suggest using principal axis analysis, this is the standard rotation factor analysis (in SAS, I use Promax).
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics