Dear Khizra,

A suggestion to begin with : Since you are drawing survey questions from studies that have not been validated, I would recommend you to do a pilot study (preferrably with a moderate sample size, say 50) and completely validate your instrument.

The main reason for choosing a sample of particular size is to capture the variance in the population. This will help us in generalizing the results to the population. I would recommend you to estimate the population size (if possible) and derive your sample size from that (See the file attached).

Next, coming to your question about smaller samples. One golden rule related to sample size is that 'larger the sample size, the better it is'. So, the ratio of ten responses to one item is not exactly correct. You may work with smaller samples also provided you are convinced about its accuracy and representativeness.

There are certain guidelines which people follow for the sake of statistical convenience ( For example, it is recommended to have a 5:1-10:1 item-to-response ratio while perfoming exploratory factor analysis). But remember these are just thumb rules!

Srikanth

Dear Khizra,

A suggestion to begin with : Since you are drawing survey questions from studies that have not been validated, I would recommend you to do a pilot study (preferrably with a moderate sample size, say 50) and completely validate your instrument.

The main reason for choosing a sample of particular size is to capture the variance in the population. This will help us in generalizing the results to the population. I would recommend you to estimate the population size (if possible) and derive your sample size from that (See the file attached).

Next, coming to your question about smaller samples. One golden rule related to sample size is that 'larger the sample size, the better it is'. So, the ratio of ten responses to one item is not exactly correct. You may work with smaller samples also provided you are convinced about its accuracy and representativeness.

There are certain guidelines which people follow for the sake of statistical convenience ( For example, it is recommended to have a 5:1-10:1 item-to-response ratio while perfoming exploratory factor analysis). But remember these are just thumb rules!

Srikanth

My advice if you're looking at non-validated questions would be don't dive right into statistical validation! Almost no surveys have actually checked what questions mean to participants. Look at options such as cognitive interviewing or even doing small qualitative assessments on people and experts that have used the tools. then, once you've verified the items actually mean what you claim they mean you can start looking to correlate them with some external criterion.

In general however, the fact you are using existing tools makes no difference to the power requirements of statistical tests, nor the fact larger samples give more accurate estimates. So I'd stick with the recommendations of the rules of thumb if I were you. Though you can consider alternative tests to the norm that may be less greedy in terms of participants required. For example, factor analysis for convergent and divergent validity is very greedy in terms of sample size and will almost certainly provide at least some spurious correlations. Cambell and Fiske have an old method of doing it which has fallen out of favour. However I think it is conceptually and computationally simple and worth considering if you wont have the power to perfom some of the more involved methods: http://psycnet.apa.org/journals/bul/56/2/81/

Dear Srikanth,

Thankyou for your answer.It is very helpful.The PDF attached was extremly helpful to calculate my sample size.Indeed my rough estimation corresponded very nearly with the calculated formula.

I had another question to clarify which is

"When we carry out Exploratory factorial analysis for validating the instrument ,will I be able to get my cronbach's alpha.I was thinking to measure the reliability of the instrument. If i could state the cronbach's alpha for internal consistancy then i could avoid administering the survey again to another sample of the same population.

My population is consisting of 410 individuals.My sample size is 220.After validation of the instrument,I would like to administer it to the remaining population inorder to know their attitude.

Is this possible?Will highly appreciate any material on this topic.

Thanks Steven for your answere.I was not able to access the article in the link sent.I would appreciate it if you were able to provide it.I would like to read on the method proposed by cample and fiske..

Dear Khizra

You are welcome :)

Yes, it is possible to report cronbach alpa for the same sample. Infact, you have to report the alpha scores for the sample used for validating your instrument.

I have enclosed a ppt. It talks about estimating alpha for identified factors (not very clear, but should be sufficient).

For more clarity, I have listed down the steps you should follow while using SPSS :

1. Perform exploratory factor analysis

2. Save the factor scores as seperate variables ( tick this option while performing FA)

3. Identify the specific items related to each of the factors.

4. Perform reliability analysis for each of the factors (using the identified items for each of the factors) seperately.

Srikanth

Forgive me for not uploading directly, but I'm in a bit of a rush. Here's another URL: http://data.psych.udel.edu/abelcher/Shared%20Documents/9%20Research%20Design%20%2815%29/Campbell%201959.pdf

If you look in google scholar for the title of the article though there are many free versions available.

Best of luck!

Steven