Consider a measurement scale with which you are familiar that you know to be a ratio scale, say, distance. If two roads are five kilometres and three kilometres long respectively then they differ in length by two kilometres. If two other roads are nine kilometres and 7 kilometres long then they also differ in length by the same amount. Furthermore, that two kilometre distance is the same in both cases; it would take the same effort to travel that two kilometre distance on either road. This implies that distance is a ratio scale.
However, suppose persons A and B score 110 and 125 respectively on some test, a difference of 15 test points; meanwhile C and D score 75 and 90 on the test, again a difference of 15 test points. In general it would be impossible to know whether the difficulty in going from a score of 75 to 90 would be the same as going from 110 to 125. But of course the scores are ordered by difficulty. Therefore, the test scores make ordinal data.
ratio
Kruskal-Wallis H test.
Ordinal. Tests responses are usually correct or incorrect. This would be assigned a value and the number of correct answers is the score of the test. There is a logical order, a correct answer is better than an incorrect answer, so it is not nominal data. Even though we calculate averages, test responses are not interval data, as there is no meaning to the interval. See related link.
The two samples must be independent and the data must be at least ordinal. Under those conditions the Mann-Whitney U test can be used.
spearman rhos
ratio
Kruskal-Wallis H test.
ratio
The independent variable in ANOVA must be categorical (either nominal or ordinal). The dependent variable must be scale (either interval or ratio). However, it is possible to recode scale variables to categorical and vice versa in order to perform ANOVA. While this is a common practice in many social sciences, it is controversial. I have also seen studies where ordinal data is treated as scale in ANOVA. Personally, I do not endorse either practice as they are tailoring the data to fit the test instead of the proper method of selecting a test that fits the data.
The Kruskal-Wallis test should be used when you have three or more independent groups and want to compare the medians of non-normally distributed data. It is a non-parametric alternative to the parametric ANOVA test and can be applied when the assumptions for ANOVA, such as normality and homogeneity of variances, are violated. The Kruskal-Wallis test is particularly useful when working with ordinal or skewed interval/ratio data.
Ordinal. Tests responses are usually correct or incorrect. This would be assigned a value and the number of correct answers is the score of the test. There is a logical order, a correct answer is better than an incorrect answer, so it is not nominal data. Even though we calculate averages, test responses are not interval data, as there is no meaning to the interval. See related link.
scatter plot
Ordered data is pretty simple. It is data which is measured (or found from a test) in ordinal types of measurement. E.G. 1 cat, 2 cats, 3 cats etc.
A F-ratio test compares 2 variances and tell if they are significantly different. A Chi-square test compares count data.
If the two distributions can be assumed to follow Gaussian (Normal) distributions then Fisher's F-test is the most powerful test. If the data are at least ordinal, then you can use the Kolmogorov-Smirnov two-sample test.
Culver Academies test scores, student-teacher ratio, parent reviews and teacher.
recess does improve test scores.