Dear researchers wherever you are,
I have some doubts about basic test. To be precise about differences in eating habits between boys and girls. In total population, 731 boys 830 girls eat and five times per day. On the contrary, 2266 boys and 2169 girls eat less than five times. Is it statistically correct to compare these groups, since in bigger one with aggregate more groups? And should we conclude with p
A chi-square test indeed gives the test of the association between eating habits and gender. In your case the association is very sign.:
Pearson's Chi-squared test with Yates' continuity correction X-squared = 718.49, df = 1, p-value < 2.2e-16
See the R script in attachment.
The boys with eating habit 5 times are overrepresented .
Dear Aleksandar
You may perform a chi2 test with the sample you have. Results will indicate a Pearson Chi2(with 1 degree of freedom) of 6.7 with a p-value of 0.01. The percentage of girls (27.7%) that were eating 5+ a day were higher than the percentage of boys (24.7%). So, you may conclude that significantly more girls than boys were eating 5+ a day. In your case, the different group sizes will not influence the results. However, the chi2 test is known to be pretty sensitive to sample size, i.e. as sample size gets larger you will achieve statistical significance very quickly. Instead of hunting p-values, it may be a good idea to think about effect size. Ask yourself: Is the difference of 3% between boys and girls really relevant? Of course, the answer is depending on your research question.
Bests, Thomas
Dear
Thomas Volken, and
Dear Jos Feys , Thank you for helping me. I have to explain my colleagues that big sample has some issue.
While big samples are not an issue per se, the test is sensitive to sample size. Check out the graph which shows p-values from a chi2-test over increasing sample size while the relative counts are maintained. The simulation started with an inital sample size of 100 per group and relative frequencies of 90/10 vs 85/15.
Chi square test is most applicable for bio statistics . Statistical relevance can be well proved .
Mr. Paul Yarnold , thank you for comprehensive contribution.
If I understand it properly, you advocate idea that chi-sqare is not best solution for this situation where on of the variables is ordinal one? I have just read an article by Prof. David Howell about the same issue. https://www.uvm.edu/~dhowell/methods/Supplements/OrdinalChiSq.html
Thank you Paul.
I am going to read more about ODA. I'm very grateful for helping me.
- Badges
- Science method