Choosing a test (1 hour)

↵ Back to module homepage

In this module we have discussed t-tests. You probably have noticed that there are other modules about other kinds of tests (such as regression and analysis of variance), and you may have other of still other kinds of tests that are not addressed in this subject (such as chi-squared tests). How do you decide what test to do? Why did we use t-tests, instead of other ones, for the examples in this module?

To understand how to choose a test, you have to understand the variables in your research design. A variable is some thing that you measure or manipulate in your study. Things that you measure might be stuff like people's proficiency, test scores, brain responses, reaction times, age, etc. Things that you manipulate might be things like which kind of training each person gets, which kind of sentence they read in an experiment, etc. There are two important distinctions to consider with your variables: whether a variable is independent or dependent, and what kind of scale the variable is measured on.

Independent vs. dependent variables

In most research you are interested in causes and effects: e.g., you might predict that giving people a different kind of training will cause them to learn grammar better, or showing people a different kind of sentence will cause their brain to react a certain way, or having higher motivation will cause people to learn tone better, or whatever. (Note that statistical methods usually cannot prove cause and effect in observational studies—see the "Types of research designs" module. Here I'm oversimplifying.) Generally, the independent variable (自變量) is the thing that you think causes something, and the dependent variable (因變量) is the effect that you want to measure. Another way of looking at it is that, in experiments, the independent variable is the thing you manipulate (either directly, by, e.g., assigning people to different training groups; or indirectly, by, e.g., making sure that your student sample has a good mix of high-proficiency, low-proficiency, and medium-proficiency students), whereas the dependent variable is the thing that you measure. The independent variable is "independent" because you choose it (you decide what kinds of groups you will look at, what kinds of words you will put in the experiment, etc.), and the dependent variable is "dependent" because it is determined by the other variables (e.g., in a reaction time experiment, you don't choose how fast a word will be responded to; it gets responded to faster or slower based on whether it's a noun or a verb, whether it's preceded by a related or an unrelated prime, or whatever other independent variable you have been messing with). In the example we discussed previously about comparing the age of PolyU students and UST students, "school" is the independent variable and "age" is the dependent variable—we "manipulated" the school variable by intentionally measuring people from two different schools, whereas we just passively observe what their ages are.

Measurement scales

Variables roughly fall into three types of measurement scale. Some variables are lumped into categories, with no inherent order. For example, in a study that compares "experimental group" vs "control group", or a study that compares "French speakers" vs. "Korean speakers", there is no inherent order; there's no particular reason to call French speakers "Group 1" and Korean speakers "Group 2" or vice versa. These are nominal variables. (Sometimes they are also called categorical variables; in R, they are called factors.)

Some variables do have an inherent order, but the distances between each level are not the same. For example, imagine you have people take a survey (such as a evaluation you all take at the end of every class) and rate how satisfied they were, with choices such as "Very dissatisfied", "Somewhat dissatisfied", "Neutral", "Somewhat satisfied", and "Very satisfied". Or imagine you give participants a language background questionnaire and ask how often they use English, with choices like "Never", "Sometimes", "Often", and "Always". These have a clear order ("never" is less than "sometimes", "sometimes" is less than "often", etc.). But the distance between levels may not be the same: maybe "often" is a lot more than "sometimes", but "always" is only a little bit more than "often". Or imagine people run a race, and you record who was first place, who was second place, who was third place, etc. There is an inherent order there, but the distances between each person may be very different; maybe the first-place runner just barely beat the second-place runner in an exciting last-minute sprint to the finish, but the third-place runner was very far behind both of them. This kind of variable is called an ordinal variable.

Finally, some variables work like true numbers. They have an order, and the distances between each number are consistent. For example, if you measure how tall someone is, or if you measure how quickly someone can finish reading a sentence, the results you get are actual meaningful numbers. This is called an interval variable (sometimes called a numeric or continuous variable). (As you learn more about statistics you may also sometimes hear about a distinction between interval and ratio variables: interval variables are numeric but don't have a natural zero [this would be things like TOEFL scores] whereas ratio variables do have a natural zero [things like length or reaction time]. In practice, this this distinction rarely has any practical consequences on what I need to do for a statistical analysis, so I usually just ignore it.)

Putting it all together

The reason this stuff is important is because it determines what kinds of test you use. Different tests are suited for different kinds of variables. For example, a t-test works great when you have a nominal independent variable (with only two levels) and a numeric dependent variable; you can review the examples from throughout this module and see that they all fit this criteria. If you have a continuous independent variable and a continuous dependent variable (e.g., if you want to see if people with higher language proficiency have faster reaction times—proficiency and reaction times are both usually measured as continuous/numeric/interval variables), a t-test will be no good. Likewise if you have a nominal independent variable and a nominal dependent variable. Likewise if you have a continuous dependent variable and a nominal independent variable with more than two levels (e.g., if your research project has three groups instead of two). For situations like that, you'll need different tests. As you've seen in previous tasks, the nature of the relationships between groups (e.g., paired vs. independent) will also influence what kind of test you need. https://stats.idre.ucla.edu/other/mult-pkg/whatstat/ is a great resource for seeing what kind of test you need in any situation.

You can find further reading on variables types, in any intro statistics textbook or online tutorial, but what was presented here should be enough to at least give you the foundation you'll need to do the remaining data analysis modules in this class.

Describe one kind of research situation that an independent samples t-test would be appropriate for, one kind of research situation that a paired samples (or one-sample) t-test would be appropriate for, and one kind of research situation that would need some other test instead of a t-test.

When you finish this activity, you are done with the module (assuming all your work on this and the previous tasks has been satisfactory). However, you may still continue on to the advanced-level task for this module if you wish to complete this module at the advanced level (if you're aiming for a higher grade or if you are just particularly interested in this topic). Otherwise, you can return to the module homepage to review this module, or return to the class homepage to select a different module or assignment to do now.


by Stephen Politzer-Ahles. Last modified on 2021-05-14. CC-BY-4.0.