If your analyses treat the variables as ordinal (i.e., non-parametric tests), then the types of scoring you described will not make a difference because they both reflect the same ordering of the scores.

Alternatively, if you want to combine the variables into a scale, then the most common approach would be to use the scoring that goes from -2 to +2.

If your analyses treat the variables as ordinal (i.e., non-parametric tests), then the types of scoring you described will not make a difference because they both reflect the same ordering of the scores.

Alternatively, if you want to combine the variables into a scale, then the most common approach would be to use the scoring that goes from -2 to +2.

I agree with David Morgan on the first issue. However if you want to combine several of these variables (with the same ordered categories) to create a new variable (or scale) pay attention that negative scores can be compensated by positive scores, so that you could have a zero total score on the new variable both in the case of all zero scores and also due to a compensation among negative and positive scores. This is the main reason for which I prefer strictly positive (i.e. greater than zero) scores.

@Daria

What's the problem of having a zero score on a scale? A 2 in a 1-5 scale "compensates" the value above 3 just the same as a -1 would "compensate" values above 0 in a -2 - 2 scale. 

If it feels like the subject didn't achieved anything then you are biased on the perception of your scale ;)

It's the interpretation of the value that matters.

@Alexander. This is definitely true, since scores should be in any case treated as interval variables. In this regard they are invariant to translation. But consider the case you want to combine scales into a new variable using multiplicative functions, which compensate less than the sum (or mean) but are sensitive to the presence of zeros and signs...

 Summing a set of variables that runs from -2 to +2 versus 1 to 5 will yield exactly the same thing because one version simply adds (or subtracts) a constant to produce the other. Hence, they will be correlated at 1.0

Thanks a lot to everyone for your answers. It helped me a lot to have enough criteria to select the scale. Additionally, I have to include scale values for "missing values" (no answers or two selected options during the questionnaires); in that case, I defined additional values, but marked them as "missing values" so will not affect other analysis (sum, average, etc.)