Fixed effects vs. Random Effects

Hi we had 5 investigators 1)hand out 100 surveys and 2)record a certain feature of each of the 100 subjects. That feature that they recorded is the outcome and may be subject to the differing biases of the investigators (eg., an investigator may be too strict all of the time in their recordings). The predictors are the survey questionnaire variables. We want to adjust for the variance introduced by the different investigators. Should we model the investigator as a fixed effect or a random effect?

(I am guessing that if all measurements were coming from the investigators I should use random but in this case only the outcome is being sampled by the investigators, the rest of the variables are being answered by the subjects.. I never came across something like this..)