Assume that there are two variables: Age groups (i.e, 1 for 18-25, 2 for 26-40, 3 for 41-50, 4 for 51-60 and 5 for 61 and above) and monthly individual expenses of 200 respondents who are evenly distributed over groups. We would like to examine these two variables together.
The question is that Which one below is more appropriate to apply:
a) Spearman correlations by accepting age groups as ordinal variable or
b) ANOVA by accepting that each age groups are independent.
The answer is important because most of the scientific commitee acepting both.
Hello Sabri. Do you have each person's actual age, or only the age group they belong to? If the former, then I suggest you start by looking at a scatter-plot with X = age and Y = expenses. If the plot suggests that the functional relationship is linear, fit a simple linear regression model. If the plot suggests some other functional form, it may still be possible to fit an appropriate OLS model (e.g., by including one or more polynomial terms).
If all you have is the age-group each person belongs to, you could use polynomial trend analysis. If you do a Google search on , you'll get lots of resources showing how to do it. For example:
- https://www.discoveringstatistics.com/repository/contrasts.pdf
HTH.
Dear Sabri,
There is a middle option: a nonparametric test that looks for trends between groups. This is known as Jonckheere Terpstra. It is available in SPSS. The problems with ANOVA are its assumptions and it doesn't look for trends. The problem with Spearman is you don't have many groups so the data is quite coarse.
Hello Sabri. Do you have each person's actual age, or only the age group they belong to? If the former, then I suggest you start by looking at a scatter-plot with X = age and Y = expenses. If the plot suggests that the functional relationship is linear, fit a simple linear regression model. If the plot suggests some other functional form, it may still be possible to fit an appropriate OLS model (e.g., by including one or more polynomial terms).
If all you have is the age-group each person belongs to, you could use polynomial trend analysis. If you do a Google search on , you'll get lots of resources showing how to do it. For example:
- https://www.discoveringstatistics.com/repository/contrasts.pdf
HTH.
@Bruce Weaver is right. On top of the reasons he gives, you should not lose potentially valid info if you have it (the exact age). But if you don't have it, a) whether the variable is ordinal or not depends to some extent on your theory: do you expect each age group to have a larger/smaller effect than the former? For many research questions in my field (voting behaviour, especially for turnout), we expect u-shaped effects; b) do you want to know if there are significant differences among in expenses among groups, do you also want to know the means /the differences in the means among age groups, or do you want to estimate the effects of age upon expenses holding constant other variables? If you add some additional info about your data and your goals, RGaters will be able to give u more precise feedback. Regards
I should add that statistical analysis should be based on the hypothesis in question, otherwise it is meaningless. If your hypothesis is about comparing the means (checking if they are equal, potentially pairwise) you should do ANOVA with posthoc comparisons.
If your hypothesis is about a relationship of those means (like testing a linear/quadratic/polynomial trend) then you can accomplish this with an appropriate contrast in ANOVA, or as described by Bruce, by using a polynomial regression fit.
- Badges
- Science topic
- Similar topics
- Medicine
- Epidemiology
- Age Groups