Login

Welcome, Guest. Please login or register.

March 29, 2024, 10:33:47 pm

Author Topic: Which statistical test of inference to use  (Read 3758 times)

0 Members and 1 Guest are viewing this topic.

Bri MT

  • VIC MVP - 2018
  • Administrator
  • ATAR Notes Legend
  • *****
  • Posts: 4719
  • invest in wellbeing so it can invest in you
  • Respect: +3677
Which statistical test of inference to use
« on: July 29, 2019, 05:32:46 pm »
+6
Two-sample t-test
“Are the two populations the samples come from different?”
This is a parametric test, meaning that it assumes the data is (roughly) normally distributed. Generally, if you have a large enough sample you can assume it will be normally distributed. If you cannot make that assumption, you need to use a non-parametric test. If you’re unsure if your data is close to normal, you can always graph it to see if it shows the “bell curve” (close to normal) or if it shows skewness (not close to normal).  The dependent variable should be measured on an interval or ratio scale.

Paired vs unpaired
Use a paired t-test if you had a matched participants or repeated measures experimental design. In matched participants, each participant is in a pair with their match, and in repeated measures, both sets of data from each person is matched. Paired t-tests have greater statistical power than unpaired t-tests, so it’s good to use them rather than you can.

Use an unpaired t-test in independent groups experimental designs. 

Assumptions for the t-test
- the data points are sampled independently
- the dependent variable is normally distributed
- the variance of the dependent variable is equal in both the experimental and control groups/conditions


Mann-Whitney U test
“the weaker, non-parametric version of the unpaired t-test”
This is a non-parametric test, meaning that the data does not have to be normally distributed. If it is normally distributed, you should use the t-test instead of this one as the t-test has greater statistical power. The dependent variable can be measured on an ordinal, interval, or ratio scale

Assumptions
- data points are sampled independently and the two samples are independent
- the data of the two samples should have the same shape when graphed


Wilcoxon signed-ranks test
“the weaker, non-parametric version of the paired t-test”
This is a non-parametric test, meaning that the data does not have to be normally distributed.  If it is normally distributed, you should use the t-test instead of this one as the t-test has greater statistical power. The dependent variable can be measured on an ordinal, interval, or ratio scale

Assumptions
- Each pair is chosen randomly and independently
- The graph of the differences between the pairs is symmetrical



If you're not sure how to use this in scientific reports/posters or have any other questions or comments please feel free to reply! :)
To do so, you'll need to register here for FREE to ask any questions you may come across in your QCE studies!
« Last Edit: July 29, 2019, 05:39:44 pm by Bri MT »