How To Choose Statistical Tests in Psychology

its not as scary as you think

Many students I have tutored, especially when they are still quite new to the research and statistical world, feel stressed as soon as they hear the word ‘statistics’.

This is the main reason why people struggle with statistics. In psychology we call it maths and stats anxiety.

Therefore, we wanted to break down how to learn statistics to help you in your journey

This certainly is not a comprehensive guide, but it should give you some building blocks to help you start in the right direction in the world of stats

1) Put the theories aside : practice > understanding

Although the theories behind the statistical tests are important to understand, their applications are far more important.

If students simply practiced performing the statistical analyses again and again, they would improve in knowing when to choose each and hence understand more clearly.

Simply put, the more you practice, the better you understand.

2) know the basics first

There are three primary types of statistical tests that you will learn in undergrad, including correlational, comparing means and frequencies

Correlation tests measure the relationship between two continuous variables. This means this type of test compares 2 variables that are each on a scale (eg IQ and height in cm; not that we’d likely find any relationship between these variables!). The most common correlation test that has been used is Pearson’s Correlation.

Comparing means tests measure differences in a continuous variable (eg score on a test) based on categorical variables. Categorical variables are variables that are split into different groups (eg male or female) or conditions (eg following meditation or prior to doing meditation). You are then interested in seeing whether these groups or conditions differ in the continuous variable.

T-tests are conducted when there are 2 groups or conditions and ANOVA is conducted when there are 3 or more groups or conditions.

If you choose t-test, there are 2 options.

One is for when the same people do 2 conditions (eg if you recruit 50 participants and they all are tested in a dark room and in a light room). This is called a paired samples t-test.

The second is if you have 2 groups (eg if you recruit 100 participants and 50 are tested in a dark room and 50 in a light room). This is called an independent t-test.

Frequencies tests are used when you are counting the number of occurrences of various categories or classes. For example, you may be interested in how many times boys and girls were late to school on a particular day or not. You would then create a table where you see how many boys were late that day and how many weren’t in one column. In the next column would be the number of girls that were late or not. The most common frequency test is the chi-squared test.

This may sound overwhelming now, but once you practice it, it gets easier

3) break the question down

Now that we know what the types of tests are break down your question.

What are your variables? Write down one line for each variable

Then write whether it is categorical, continuous or frequency data next to each

If you have two continuous variables, you know you’re looking at correlations (Pearson’s Correlation)

If you have frequency variables, you will do a frequency test (Chi-squared test)

If you have a categorical variable and a continuous variable, you are looking at comparing means

In the case of compare means, you will have to ask a couple more questions

How many levels are there in the category? If 2 its a t-test, if 3 or more it’s ANOVA

If it’s a t-test, are the same people doing different conditions? If so, it is a paired samples t-test

If different people are in different groups, you would do an independent t-test

Now, practice practice, practice

Here are some example questions (answers are at the bottom of this blog post)

1) the effect of therapy vs no therapy on Rosenberg self esteem scores (people can score between 0 and 40 on this measure). Out of 100 participants, 50 were randomized to the therapy group and 50 were randomized to the control group.

2) we counted the number of males and females who completed a degree following school. We then compared whether there was a difference between them in the frequency of completing a degree

3) participants self esteem was calculated on the Rosenberg self esteem scale (people can score between 0 and 40 on this measure) and were asked how happy they were on a scale of 0-100. The researchers were interested in the relationship between these two variables

4) male/ female/ non binary were compared on measures of height (in cm). The researchers were interested to see whether gender group was related to height

5) out of 100 children, all 100 performed 2 conditions. Participants were placed in rooms and presented with a marshmallow. They were told that if they waited 10 minutes for the experimenter to return they could eat 2 marshmallows. In the first condition participants were tested on how long they could resist the marshmallow (in seconds) in a small and cramped room. In the second condition, participants were tested on how long they could resist a marshmallow in a larger and more empty room. Researchers were interested to see whether condition (whether it was a small or large room) had any effect on the time participants lasted.

4) knowing the assumptions of each test (for once you’ve nailed the basics)

Sorry but here it becomes a bit more complex.

In some circumstances, you will not be able to do the primary statistical tests mentioned in points 2 and 3.

This is because sometimes your data doesn’t meet assumptions that are needed to do that test

Assumptions are like a checklist which makes sure you meet the criteria for the gold standard tests (Pearsons correlation, t-tests, ANOVA or chi-squared).

If you don’t, you have to choose the equivalent which doesn’t require assumptions.

Eqiuvalent of a Pearsons Correlation is Spearman’s Roe

Equivalent of an independent t-test is Mann-Whitney U

Equivalent of a paired-samples t-test is Wilcoxen test

Equivalent of ANOVA is Kruskal-Wallis

Equivalent of the Chi-Squared test is Poisson regression

We do not discuss each of the assumptions in this article as that would end up very long, but if you’d like to learn more about them, you can go here

If your data doesn’t meet the criteria for an ANOVA for example, you would choose the Kruskal-Wallis test.

5) ask for help if needed

When you have a statistical assignment, try to break down your assigned study variables as much as you can to work things our for yourself. If you can’t understand, try to talk to your lecturer or your tutor at school or university. After that, you should try out different statistical tests by using the data provided (or collected) for your assignment. There is nothing wrong to try out different statistical tests before you choose, because this can help you differentiate the statistical tests, which can also help you to choose the right tests for your analysis.

If you want some specialist help in learning how to approach statistical tests, we’re here for you.

Answers

1) Independent t-test

2) Chi-squared test

3) Pearsons correlation

4) ANOVA

4) Paired-samples t-test