I have done a short survey and until now, I was able to collect only 100 responses. Not sure if collecting more would be possible. So was wondering, what kind of statistical analysis would support a sample size of 100?
Or if you have any other suggestions, please provide.
Hello Somya,
The answer depends on your specific research question and the nature of the variable(s) you will be evaluating. If you could elaborate on these, you'd be more likely to get helpful responses.
The "384" sample size recommendation comes from typical formulas for estimating population percentages/proportions from sample values (from simple random samples) with a certain precision (+/- 5%) and certain confidence level (95%). You can always estimate parameters with more or fewer cases than that, but the resulting confidence interval will be narrower or wider, respectively. If this isn't what you're doing, then the value isn't germane to your situation.
Good luck with your work.
Hi Prof.Somya,
To analyze data with a small dataset you can use statistical techniques that are not derived from the OLS model. In Statistics, we call it non-parametric statistics. For example using spline technique. However, you can also try methods which estimate parameters using Bayesian or also Partial least square. At the same time, to increase accuracy, you can try various combinations of methods + optimization. For example, Bayesian Regression + MCMC.
Its depends if its infinite population say for example (Generation z, IT employees etc) then the thumb rule is 384 (as per morgan sample size determination)
In your case please share more details on who is your respodents!
Thanks and regards,
Gandhi Vasanth Kumar
Gandhi Vasanth Kumar thank you for your reply. I am doing a survey on international people living/ studied/ worked in Taiwan. So, students, people working in different organizations, alumni are my respondents.
Rezzy Eko Caraka thank you for your suggestion. I will have a look at these methods.
Gandhi Vasanth Kumar so you mean to say, I need to have at least 384 responses? But that seems a little difficult. Any idea if I can still do SEM analysis with this data? Or any other analysis using SPSS?
Hello Somya,
The answer depends on your specific research question and the nature of the variable(s) you will be evaluating. If you could elaborate on these, you'd be more likely to get helpful responses.
The "384" sample size recommendation comes from typical formulas for estimating population percentages/proportions from sample values (from simple random samples) with a certain precision (+/- 5%) and certain confidence level (95%). You can always estimate parameters with more or fewer cases than that, but the resulting confidence interval will be narrower or wider, respectively. If this isn't what you're doing, then the value isn't germane to your situation.
Good luck with your work.
Hi Somya Agrawal
It' really depends on your objectives, type of variables (categorical or continuous), type of analysis and how much groups you have.
Basically, t-test, ANOVA and Pearson correlation it's more common. If you have suspicious of normality try nonparametric tests like Mann-Whitney, Kruskal-Wallis and Spearman correlation.
In same way, think in Latent Class Analysis (poLCA R-package) or Latent Profile Analysis (tidyLPA R-package) as a alternative to SEM analysis.
In some cases 100 answers it's enough.
David Morse and Israel Souza thank you for your responses. I'll try to explain my situation. I have in all 10 latent variables (having 4 items each) which includes the independent and dependent variables. Along with this, I have a few demographic (dummy) variables such as gender, age, home country, etc etc. As of now, I have collected around 150 responses. Initially, my plan was just to study the interrelationships between all the variables using Hierarchical Regression Analysis (SPSS) and also maybe do Structural Equation Modelling (AMOS). But I am not quite sure whether a sample size of 150 is good enough to compute the analysis to extract some interesting results. Hence the question on what type of analysis would be more suitable for a sample size of 150 in order to be able to build a good story around the variables. I also have some open-ended questions which I can use as well. Your suggestions are still welcome! :)
Hello again, Somya,
Your aim sounds as if SEM (structural equation model/ing) is the appropriate tool. The glib advice for sample size in SEM models is 10-20 cases per parameter to be estimated. So, if you have a proposed model you'd like to test, just count the number of unknown values to be estimated and go from there.
A little more insight could be gained by trying:
1. an online calculator such as this one: https://www.danielsoper.com/statcalc/calculator.aspx?id=89
2. or, if the outcome fit index is important, this one: https://timo.gnambs.at/research/power-for-sem
3. simulations, using approaches such as this in the R system: https://bookdown.org/marklhc/notes/simulation-example-on-structural-equation-modeling-sem.html, or
4. a pilot study, using a small portion of your current data.
Good luck with your work!
Hello Somya Agrawal , I agree with David Morse , if the latent variables is good enough and your data have good structure. Sometimes, despite recomendations, trial and error method is a fundamental key.
Pay atention to residuals and error level.
Best regards
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis