I am analysing the results from a survey we conducted about the awareness of genetic testing. I am using semantic differential scale. After applying one way ANOVA, there is a significant difference in the level of awareness, depending on the different source. However, after applying Spearman correlation test, there is a negative correlation, but it is not significant. I also did linear regression analysis, but R squeare is 0.04. How should I interpret this data?
From your description, it is not clear what your design is (within or between ANOVA?? what do you mean with different source?). Why Spearman, if you have any doubts that you have an interval scale or that your data/residuals are not normally distributed, then ANOVA would not be appropriate, too. Why ANOVA, if you have only two groups or conditions (which I take from your regression account), a t-test would suffice (although the same statistically speaking). Measuring differences and calculating correlations is not the same for within/repeated measures designs!!! For example, take one variable, add constant to it and calculate a within t-test, you would have a perfect correlation (since they are invariant to linear transformations) but you would find a difference in the means. In a between design with two groups the pointbiserial correlation, regression with dummy variables and t-test should be equivalent.
I could go on with some other things, but I am really not sure what you did, what you want to measure and if you know what you did in the first place.
Rainer Duesing Thank you for your answer. I am also very confused. I am checking if the level of awareness of genetic testing on the scale from 1 to 10 depends on the source of information. For example, I expect that people who received some information from their doctor would know more compared to people, who searched for information on the internet. Maybe a t-test is more appropriate than ANOVA. I used Spearman to search for a correlation between the different source of information and the level of awareness.
So, you have two independent groups which are mutually exclusive as a IV? Go with t-test. But why did you choose Spearman correlation, what was the reason not to use the typical Pearson correlation. Spearman might be troublesome, because ties in the ranks can lead to problems in the calculation/interpretation and here you have a special case with only two values in your IV. I cant predict what happens there, but maybe you should simply try the Pearson correlation and see what this will tell you.
On the other hand, Spearman is the correlation with the ranks of the data, so that it should also work with binary data. Can you show your data please?
So, I checked, no problem in SPSS and R with fake data, both all analyses, ANOVA, t-test, Pearson Correlation, Spearman Correlation and linear regression come to the same conclusion (as expected), so maybe you did something wrong? Please share your data structure.
Alright, I think you really did not look into statistical methods enough! Apparently, your IV "source" has 4 levels, which are nominal and mutually exclusive, right (you can be either 1,2,3,or 4). Therefore, ANOVA approach would be correct in principal (although your residuals are not normally distributed, so that I think ANOVA is not suiteable), but of course not any form of bivariate correlation, since your "source" variable is nominal and not ordindal or interval scaled. Applying a correlation makes absolutely no sense, since the numbers for the different sources are totally arbitrary, e.g. instead of naming the categories 1,2,3,4 you could have named them 2,4,3,1 or 101,2,1004,99, it makes no difference for the ANOVA, but the correlation would be dependend on that labeling, since it will take it as values and not strings.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA