I am a master student and writing research paper. I have 3 yes/no questions, 2 multiple responses questions and 1 Likert scale question. I would like find out whether age has relationship to my variables. I grouped age in my questionnaire such as 18-30, 31-45, 46-55. Can I use chi-square test?
Thank you very much.
Chi-square test is commonly used for analyzing categorical data. So, yes you can use it to analyze the relationship between age groups and your variables.
On Ki Why treat age as a discrete variable rather than a continuous variable? Instead of chi-square, I would treat age as a continuous variable and use logistic regression instead of chi-square.
adding to Blaine's answer: the analysis of the Likert-scale response would then be done using an ordered logistic regression model.
Also note that age can be modeled quite flexibly, e.,g. over a spline basis, to model arbitrary non-linear relationships between age and the expected response.
And finally, note that testing coefficients in logistic regression models (which are log odds ratios) will be done based on the chi²-approximation, so these are chi²-tests, in the same way as testing the slope in a linear regression model is based on the t-distribution and therefore is a t-test.
I'm going disagree somewhat with the previous answers.
The first question is, Do you just want to find if there is an association between the pairs of these variables ?
Or do you want to construct a larger model with all these variables ?
If so, which variable is your dependent variable ?
A multiple response question is usually treated as multiple yes / no questions. That is, each checkbox is a separate yes / no question.
If you want to find if there is an association between each pair of variables, you can just use Kendall or Spearman correlation.
A dichotomous variable can be treated as ordinal. It's the same either way. So all of your yes/no variables, Likert-type item, and age categories could be treated as ordinal variables.
Sal Mangiafico By separating age into four, discrete bins some of the age-related variance is removed, thereby losing information. The risk of doing this is one might observe a significant difference among two or more of the age groups on the DV when age is treated as discrete categories, but no significant relationship between age and the DV exists when age is treated as continuous. If this is the case, we'd mistakenly conclude age is related to the DV, when it really isn't.
Blaine Tomkins . I don't disagree with that point.... The problem is all these answers are assuming what the DV is, and what OP is trying to do...
Agreed. Some additional information about these items/questions would be helpful. I just assumed it was more intuitive to use age as the IV, in which case the situation calls for logistic regression - at least for the dichotomous item and polytomous items.
On Ki - do you plan to combine these questions into a single score or analyze them separately?
As I understood, there are two kinds of variables: Answers (binomial or ordinal) and Age (continuous). To me it seemed natural to ask: how do the probabilities of giving certain answers depend on (or change with) age? Or: given I know the age of a subject, what can I expect about the answers this subject will give? This, to me, implies that age at least indirectly affects the answer-probabilities or can be used as a surrogate variable to say something more specific about the expected answers. To me this makes more sense than using answers to get a more specific idea about the age of the subject (although this would of course be possible - but it would be so much simpler to ask for the age directly, and it is also harder to imagine that the probability of giving certain answers would affect the age of the subject...).
- Badges
- Science method