 |
 |
Dear FSL team, I am unsure as to when it is vs isn't worth including extraneous factors as EVS in one's design. For instance, say that my sample is formed of people in their early 20s, and with an even gender distribution. I am not interested in the effect of either of these factors and just want to check group differences and brain-behaviour correlations "across the board", with the aim of maximising the external validity of my inference. How can the following two contradictory (but, seems to me, equally valid) points of view be reconciled, and is one of them better to follow than the other: 1) Since age and gender clearly *do* have an effect on brain structure, these two factors need to be included as EVS in any design, even if one is not directly interested in their effect, simply to factor out variance of-no-interest that is due to them. Not doing this can lead to both Type I or Type II errors, e.g. if some of this variance of-no-interest is mistakenly attributed to factors of interest such as group or behavioural score. 2) Each added EV takes away statistical power, and so unless one is interested in the effect of e.g. age, one should not have to "pay the price" for having them in the design. It's OK to not code for nuisance variables like age and gender just as it is OK to not code for other uninteresting things, like whether someone is or isn't a smoker. In particular, it is all the more OK to not code for age if the sample's age range is reduced. I realise this may be a statistical point of contention but it'd be really helpful to hear your views on it! Anticipated thanks! Tudor