“What’s that noise?” — Of systematic and random errors, or: Why pilot testing is important

If you study social sciences, you might be especially interested in this section about types of errors.

Test theory assumes that every score or observation is composed of the following two factors: the true value of the variable plus an error. This error can either be at random or systematic. 

The random error (sometimes also called noise) is caused by factors that affect the measurement of the variable of interest completely at random. A random error can be, for example, the participants’ mood when taking an intelligence test, which may affect the performance of some participants but not of others — either in a positive or in a negative way. Random errors do not have a systematic effect on your whole sample, which is why they typically do not affect the average group score(s). 

The systematic error (sometimes also called bias), however, is a factor that influences the measurement of the variable of interest in a systematic way, that is, in the whole sample. For example, if you are administering an intelligence test and the building in which the test is taken is being renovated, then the potentially noisy environment will most likely lower the scores of all participants. As a researcher, you would really want to avoid or minimise this kind of systematic bias – otherwise your data may not reflecting what you're actually investigating!

The good news is that there are some things that you can do about both types of measurement error.

First, please do pilot test your questionnaire or instruments, in order to get feedback from your participants on potential sources of influence and bias. We generally recommend that you pilot test your study with a handful of participants. 

Second, we recommend that you also ask your colleagues / your research lab to provide you with feedback on your study. After all, that's what teams are there for! 🙂 While participants can only see the "appearance" of your study, your colleagues may in addition be able to sanity check your branch logic and report any bugs back to you. Trust us, this approach is invaluable in preventing all sorts of biases, bugs or external factors undermining your study! We've seen some studies go awry because of easily preventable bugs, so we recommend that you avoid plowing ahead with your study without any sanity checks.

Pro-tip: One possibility of minimising measurement errors is to have several methods to measure the same variable you are interested in. This step is especially useful when you know that a certain method is prone to be systematically biased. 

Third, assuming you have pilot tested your study and collected all the data you wanted, you should double-check that you did not make any mistakes when importing / handling / merging your data. Sounds obvious, but you'd be surprised how many stories we've heard of wrongly coded variables and responses that produced spurious results. Just ask a colleague to take a quick a look. Sanity checks like these can go a long way. 

Finally, when it comes to data analysis, there are some statistical measures that you can apply to quantify your measurement error. Besides, the point above regarding sanity checks applies to analysis code in exactly the same way! Do check your code for any errors, and ask a buddy to double-check it. All of the above will significantly increase the quality of your research.

The expertise of the Prolific Team & https://socialresearchmethods.net/kb/measerr.php

Was this article helpful?
1 out of 1 found this helpful