home | about |
links | tools### On the Accuracy of Samples^{1}

The question often arises -- "How accurate is my sample?" The
answer to this question is trickier than it sounds, since the level of accuracy
will vary from question to question within a survey. Since the variance is
different from one survey question to another, the survey's precision will vary
from item to item.

The precision of a survey depends upon the variance of the
responses to the question of interest, the level of confidence used, and the
sample size. When the sample represents a significant proportion (experts say
this can be anywhere between 2%-10%) of the population, this too must be taken into account when computing precision.

The approach often used by pollsters and researchers in
reporting the precision of nationally representative surveys is a conservative
approach; it assumes the "worst case." The worst
case is represented by dichotomous data (0-1, yes-no) with maximum variability
(50/50 distribution of responses). This approach is popular because it is
simple. The first link below uses this approach to calculate the relative margin
of error for such data. The second link below computes a measure of absolute
precision for continuous data. This requires an estimate of the mean and
variabililty of the relevant population.

Form to
calculate precision of proportion data