I do not have the expected counts for calculating the chi square as I do not have any previous study from where I can get that expected count. Though the SPSS software will calculate the expected counts for me and will do the goodness of fit analysis. Will it make sense to use the Chi Square when we do not have any previous study to refer to for the expected counts and test the data distribution for the goodness of fit based on the system calculated expected values in the SPSS. Can it lead to some issues?
Hello Amit,
To run a chi-square you don't need a previous study. The expected counts are calculated under the null hypothesis, and in my understanding you can use them in a goodness of fit test.
Best
Expected counts are the ones you compute under your hypothesis ( the model assumption) regarding the variable under study for which observations are collected and observed frequencies noted.
The expected counts are typically based on some theory (i.e., what Subrata Chakraborty called the model assumption). What is your variable, and what are the observed counts? Are there any relevant theories in the literature?
Hi @Bruce Weaver, I am having a research project in which I would like to use the goodness of fit test for the gender variables (male/female/other category). I have tried to get some information from govt's statistical department to know about the male and female numbers for the entire city (age group 18-24). I have got that expected count now from them ( Male 51 % and Female 49%. But no figure available for the other gender category). My collected research data (for gender variable) also falls under the same distribution. I want to use goodness of fit test for this case. Similarly, I have age variable(18-24 yrs- interval variable) and have got the expected count for each age level. Though, I know that chi goodness of fit is only for the categorical variable, can we have any statistical method that can be used to describe goodness of fit for scale variables (interval/ratio) or can it be best left to the reported frequencies of each category under that scale variable?
Hello Amit. Using a Chi-square Goodness of Fit test will be problematic for your gender variable, because the Other cell will have an expected frequency of 0. As Thom S Baguley says in his book Serious Stats (p. 133), "A reasonable recommendation [for the Chi-square goodness of fit test] is that Ej >= 5 for at least 80% of cells and that no cells have expected values of 0."
You listed SPSS as one of the topics. If you have the Exact Tests module, you should be able to get an exact multinomial goodness of fit test. Be aware though that SPSS will not allow you to enter 0 as the expected proportion. You'll have to enter a very small decimal instead, something like 0.0001. Here's an example I just tried on a file I had open:
NPAR TESTS
/CHISQUARE=gender
/EXPECTED=.50995 .48995 0.0001
/MISSING ANALYSIS
/METHOD=EXACT TIMER(5).
Notice that I reduced your expected proportions of .51 and .49 by 0.0001/2 so that the expected proportions would sum to 1, as they should.
It wasn't clear to me what you were asking about your age variable. Do you have the actual ages, or just the age categories?
HTH.
Thank you Bruce Weaver for your reply. Yes, the problem was there regarding the expected count for the gender variable as the Chi-square goodness of fit test was not being run due to this issue. Total sample size is 127 out of which only 2 responses belong to "other" gender category. What I did is that, I have run the goodness-of-fit for the remaining 125 responses that fall under either male or female category and have not considered 2 responses that were under "other" gender category. I wish to report this in my research thesis. I can also give the explanation that as the statistical data from the Govt regarding the "other" gender category is not available, the "other" gender category has not been considered for the goodness-of-fit test.
Regarding the Age Variable- I wanted to ask that (as my my understanding), we use Chi-square goodness-of-fit test for the categorical (nominal/ordinal) variables. Can it also be used for the "age" variable ( which at times can be ordinal if arranged in groups, continuous). In my research, the age variable has age values (18,19,20,21,22,23,24 years as the topic is based on Generation Z). Can I use Chi square goodness of fit for my "age" variable or is there anything else that should be appropriate. I have got the data form the Govt Statistical Dept regarding the total population of this age group (age-wise e.g 18,19,20 yrs..) and I want to check whether my collected data fits well with the population data ( The population representation N= 25000 APPROX FOR YEAR 18,19,20,21,22,23,24 AND THEIR % REPRESENTATION INCREASES AS THE AGE INCREASES).
Thank you so much for taking the time to clear my doubts.
Warmly,
Amit
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running