CNN 2000ats Page 132
Exhibit A.2
Approximate Sampling Tolerances (at 95% Confidence) to Use
in Evaluating Differences Between Two
Percentage Results
Approximate
Sample Size of
Survey
Survey
Survey
Survey
Two Groups Asked
Percentage
Percentage
Percentage
Percentage
Survey
Question on
Result at
Result at
Result at
Result at
Percentage
Which Survey
10% or
20% or
30% or
40% or
Result at
Result Is Based
90%
80%
70%
60%
50%
5,000 vs. 2,000
2
2
2
3
3
1,000
2
3
3
3
3
500
3
4
4
5
5
300
3
5
5
6
6
200
4
6
6
7
7
100
6
8
9
10
10
50
8
11
13
14
14
2,000 vs. 2,000
2
2
3
3
3
1,000
2
3
3
4
4
500
3
4
4
5
5
200
4
6
7
7
7
100
6
8
9
10
10
50
8
11
13
14
14
1,000 vs. 1,000
3
4
4
4
4
500
3
4
5
5
5
200
5
6
7
7
8
100
6
8
9
10
10
50
9
11
13
14
14
500 vs. 500
4
5
6
6
6
200
5
7
8
8
8
100
6
9
10
11
11
50
9
12
13
14
15
200 vs. 200
6
8
9
10
10
100
7
10
11
12
12
50
9
12
14
15
15
100 vs. 100
8
11
13
14
14
50
10
14
16
17
17
50 vs. 50
12
16
18
19
20
Non-Sampling Error
Sampling error is only one way in which survey findings may vary from the findings
that would result from interviewing every member of the relevant population. Survey
research is susceptible to human and mechanical errors as well, such as interviewer
recording and data handling errors. However, the procedures used by the Harris
firm, including the CAI systems described earlier, keep these types of errors to a min-
imum.
125
