Covariates, demographics and manipulation check
Covariates are psychological factors that could influence the data. Your experiment cannot manipulate them experimentally (think of personality). You want to measure these covariates. Demographics too have to be measured in all experiments. And last but not least, manipulation checks are usually included in experiments with manipulations of an independent variable. This article explains how to measure these variables.
Think carefully about any psychological constructs that may influence your dependent variable. Let’s say you want to find out if a new teaching style improves learning in college students. One group of students is exposed to the new teaching style, a second group is taught with the old style. After the semester you compare the groups’ performance in a final exam. Sounds pretty straightforward, doesn’t it? But wait – students’ performance is not only influenced by teaching style: test anxiety, motivation, intelligence, interest in the topic etc. can also play an important role.
We call these variables covariates. If some of them systematically differ between experimental groups, this can be an alternative explanation for an observed difference in the final exam scores. Therefore, you need to decide in advance which are the potentially relevant covariates for your study and take them into account when designing your experiment. There will probably be many potential covariates, so you may need to settle for capturing only the most important ones.
The tricky part about covariates is that often they are latent constructs. Latent constructs are usually complex and abstract, and not directly observable. To measure them we use carefully designed tests and questionnaires.
For example, if our covariate is neuroticism, you can use a suitable questionnaire from the literature, like the NEO Five-Factor Inventory (Costa & McCrae, 1992). Participants’ score on the neuroticism subscale of this test is the measure of your covariate. One construct can be measured in multiple ways, for instance, subjective pain can be measured through self-reported pain intensity, observing pain behavior (flinching, crying, avoidance of painful stimuli etc.), or a clinician’s judgment, just to name a few.
Finding the right measure for your covariate is crucial in designing your experiment. Below we show you where to look for it. Note, it is not advisable to invent and use a new questionnaire without validating it. Whereas no single measure will perfectly capture your construct, good measures should be administered in a consistent way, i.e. be standardized. They should produce replicable results, i.e., be reliable. And they should measure what they are supposed to measure, i.e. be valid. Also it is nice if they are, efficiently administered and add the most relevant information for your experiment (don’t just measure everything).
Definitely discuss potential covariates with your research team, and find the best measure – rather than the first measure – to measure the constructs of interest!
Where to find standardized tests?
If you are studying at a university, you don’t need to buy tests (many psychological tests are licensed, costing 300 USD or more due to expensive research and development processes, think of medicine or books). Here are three ways to find standardized tests: the database PsychTEST (if you have access), a test library (if you have access), or material from previous studies. The University of Basel offers access to all three.
PsycTESTS is an online database of psychological measures, scales, surveys, and other instruments. It includes aptitude tests, personality scales, cognitive functioning measures, etc. The majority of them are available for download and they include reliability and validity information.
Important: Access requires the VPN-Client or a connection with Eduroam.
The Testothek is simply a library of psychological tests. Borrow a standardized test, like a book, from there. If you work in a psychology department, check with your library if your department offers a test library, often called Testothek in German-speaking countries. If you wish, you can check out our video tutorial for the Testothek.
Previous studies can be very helpful to find testing materials for your experiment. There are three different approaches to maintain them:
➔ Appendices of papers: While looking for supporting literature for your study, you probably came across some papers that used test material you could also apply to your experiment. Have a look at the method section of these papers. Go to the part where they describe the task and you may come across something like ‘Testing materials/tests/stimuli are presented in Appendix X’. If so, you can simply go to Appendix X and find your potential measures.
➔ Citations: Some studies use measures that have been described in a previous study. Therefore, they provide a citation of the study where the measures are presented in greater detail in the method section. If so, simply look up the cited paper to find the instrument. Here is an example of finding measures by using previous studies:
➔ Contact the authors: Sometimes during your ‘test hunting’ you will find statements that the material is available upon request. If this is the case, don’t shy away from asking the authors for it. You’ll usually need to provide a brief summary of your experiment in order to get the instrument. Also, it may take quite some time for the authors to answer, so if using resources in this manner, do not leave this task for the last minute.
Demographics: Why the extra information?
The bare minimum demographic information involves a record of participants’ age, sex, and occupation. Depending on the specifics of your experiment, you can further extend the demographic records to include variables such as: educational attainment, family size, first language, marital status, nationality, religion etc. It depends on your study design!
Other variables: Manipulation check
Manipulation checks are important: Does the experimentally-manipulated independent variable vary like you expect it to? Consider studying the effect of emotions on memory by exposing participants to a sad film compared to a happy film followed by a memory task. After the film and before the memory task ask participants how sad or happy they actually feel! This safeguards against the case that observing no between-group differences is merely a result of a failed manipulation, rather than an actual lack of impact of the independent variables on the dependent ones. Without manipulation check there is simply no way to be sure about this.
Costa, P. T., & McCrae, R. R. (1992). Revised NEO personality inventory (NEO-PI-R) and NEO five-factor inventory (NEO-FFI): Professional manual. Odessa, FL: Psychological Assessment Resources, Inc.
University of Basel