As expected for an application of magnitude scaling without a common modulus,
dollar responses are statistically less efficient than category scale measures of the same
attitudes. We have seen that the averages of different attitude expressions in large
samples yield similar rankings of objects see Tables 1 and 2.. However, dollar
responses produce much lower signal-to-noise ratios than do rating scales. Tables 4
and 5 present results from separate analyses of variance for each of the response
measures used in the two studies. The analysis partitions the variance of responses
into three components: i. Object signal.: the variance associated with differences
among objects of judgments e.g., public goods that differ in value, personal injury
cases that vary in the outrageousness of the defendant’s actions.. ii. Respondents:
the variance associated with individual differences in the mean level of responses,
over objects e.g., some respondents state generally higher WTP than others, some
experimental jurors are generally more severe than others.. iii. Noise: the residual
variance, which combines the effects of individual differences in variability, idiosyncratic
responses of some respondents to some objects or topics, and various sources
of measurement error.