For example, I might ask participants to complete a visual analogue scale for pain. They would be given a line that is 100mm long, and one end says 0 (No pain) and the other end says 100 (worst possible pain). The participant is asked to make a mark on the line with a pencil or pen to rate their level of pain. I would then use a ruler to measure the distance from the origin. E.g. how many millimeters away it is from the start of the line (i.e. zero).
The problem is that many times, people do not place a mark such as an x on the line, but they might circle a part of the line, or put a tick (check mark) there instead. I attach a picture of the standard response (participant 1) followed by 5 possible scenarios where the answer is ambiguous (participants 2 to 7), so I have left it as missing data/blank. Should these be excluded from the dataset (coded as missing data) or are some of them salvageable? Possible responses to visual analogue scales
number 7

By way of explanation: In number 2) the line is very large so there isn’t one place that it intersects the horizontal line (usually it has to be accurate to the nearest mm).
Participant 3) has placed a circle on the line so it is difficult to say where on the line they meant. Participant number 4) has circled one of the numbers at the end. Can I be sure that they meant 0? Participant 7 is a variation on this.
Participant 5) has placed a tick (check) mark underneath the line (or sometimes they place a tick on the line itself). Which part of the tick do I measure from?
Participant 6) has placed a tick (check) mark on one of the numbers at the end of the scale. Can I be sure that they meant 100?
What is the best way to minimize bias while maximizing the use of the available data? Any thoughts would be appreciated.

  • $\begingroup$ Thank you very much AliceD and MariaAnt for your thoughts, much appreciated. The sample is quite large (greater than 100 but there are smaller subgroups). I agree that number 4 and 6 are the riskiest. Keen to hear any other thoughts and anyone who might have encountered this situation and how they dealt with it. $\endgroup$
    – m_moth
    Aug 19, 2015 at 11:44

2 Answers 2


I fully agree to AliceD's thoughts. If you have a large sample, these problems might add a little noise, but won't make much of a difference (certainly not the cases 2, 3, and 5, because the psychological resolution of the scale is unlikely to be as precise as the millimeter measure suggests anyway).

In addition: If you want to be sure not to introduce any systematic bias yourself, the most prudent procedure would be to code this blind to conditions (if you have any) or other measures of the study. This way you can avoid that you unwittingly help you own hypotheses.


I am not an expert on these scales, but my 2-cents worth would be the following, which are simple work-arounds to include all the data:

  1. Fine as is;
  2. Take the middle of the thick line (the 'average');
  3. Take the center of the circle (draw a cross in it, the intersection is the middle);
  4. Apparently "0" - no pain;
  5. Take the tip as the marker;
  6. Apparently "100".
  7. As in 4. ("0" - no pain);

Points #4 and #6 are obviously the most risky, especially in a small population. In a large population (say N=100) a few outliers don't matter too much. So in case the study sample is small, you may want to consider excluding uncertain cases (#4 and #6 as examples), or try to contact the subjects and re-assess.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.