How to fix the sample size for human questionnaire study
Hi,Generally we calculate the samples size based on the number of items(questions statements) in a survey question.The general principle is no of items into 10.Sample size can also be by the KMO.Read more about it.
The above answers are OK but show you different research questions (methods) require different solutions. Please describe yours in a bit more detail. Best, David Booth
Hello Sakthivel,
There are two primary approaches to answering questions about sample size.
1. For the purposes of estimating some parameter(s) of a population (either finite or infinite) with some desired degree of precision with some sort of probabilistic sampling method, a number of great sources exist for helping to determine the requisite sample size. Here's a couple of related resources:
Types/approaches to sampling:
https://towardsdatascience.com/sampling-techniques-a4e34111d808
Sample size determination:
William Cochran, Sampling techniques (3rd ed.). https://www.researchgate.net/...sample.../Cochran_1977_Sampling+Techniques.pdf
Here's a very simplified example, for a simple random sample (every case in the population has an equal probability of selection, and every possible subset of k cases--for a sample size of k--has an equal probability of selection), assuming you want to estimate the percent of an infinite population that engages in some behavior with an accuracy of +/- 5 percentage points, and to miss that degree of accuracy no more than 5 percent of the time. N = (t^2)*(Variance) / L^2.
t is the critical t-statistic for statistical significance at the .05 level ("5 percent of the time). With large sample sizes, t will be approximately 1.96 (I'll use 2 for simplicity). So, t^2 is approximately 4.
Variance is the variance of the behavior in the population. For percents/proportions of a yes/no or dichotomous variable, variance is percent*(100 - percent). Worst case (largest variance) example is for a population value of 50%: variance = 50*(100 - 50) = 2500.
L is the desired accuracy (here, +/- 5 points). So, L^2 is 25.
N = 4 * 2500 / 25 = 400.
For simple random samples, that number will be lower with: (a) a finite population; (b) lower variance; (c) higher possibility (than 5%) of being altogether wrong; (d) less desired accuracy; or (e) some combination of these factors.
2. For the purposes of determining whether a hypothesis test will have some desired level of statistical power to detect some desired effect size (if present in the population) when tested at some desired alpha level, power calculation software, such as the freely available G*Power program (http://www.gpower.hhu.de/) is quite helpful. See the user manual and short tutorial (available at the same site) for examples.
Good luck with your work!
Everything David Morse says is correct. Along with that remember Jaspreet Kaur is also correct. The Marketing Research folk use her approach. For more details see the link: https://www.google.com/search?q=Sample+size+calculation+for+marketing+research+surveys&rlz=1C1CHBF_enUS847US847&oq=Sample+size+calculation+for+marketing+research+surveys&aqs=chrome..69i57.34356j0j7&sourceid=chrome&ie=UTF-8 BTW in my opinion Cochran's Sampling Techniques is one of the best statistics books ever written. Thanks Dr. Morse for mentioning it.
BTW the KMO approach mentioned by @Jaspreet Kaur is discussed here: https://www.google.com/search?q=KMO&rlz=1C1CHBF_enUS847US847&oq=KMO&aqs=chrome..69i57j0l5.6455j0j9&sourceid=chrome&ie=UTF-8
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size