Appendix D: Statistical Analysis of Differences

All Documents -- All Issues

NFA with Checklist

NFA Without Checklist

Number found

61

62

Total issues

70

115

Percentage found

87.14%

53.91%

Difference in percentage

33.23%

Confidence level

99.9997%

CTA Documents -- All Issues

NFA with Checklist

NFA Without Checklist

Number found

43

20

Total issues

47

58

Percentage found

91.49%

34.48%

Difference in percentage

57.01%

Confidence level

99.9999997%

CTA Documents -- More significant Issues

NFA With Checklist

NFA Without Checklist

Number found

16

8

Total issues

18

22

Percentage found

88.89%

36.36%

Difference in percentage

52.53%

Confidence level

99.93%

CTA Documents � Less significant Issues

NFA With Checklist

NFA Without Checklist

Number found

27

12

Total issues

29

36

Percentage found

93.10%

33.33%

Difference in percentage

59.77%

Confidence level

99.9999%

CPO Documents -- All Issues

NFA with Checklist

NFA Without Checklist

Number found

18

42

Total issues

23

57

Percentage found

78.26%

73.68%

Difference in percentage

4.58%

Confidence level

33%

CPO Documents -- More significant Issues

NFA With Checklist

NFA Without Checklist

Number found

4

6

Total issues

4

11

Percentage found

100.00%

54.55%

Difference in percentage

45.45%

Confidence level

90%

CPO Documents -- Less significant Issues

NFA With Checklist

NFA Without Checklist

Number found

14

36

Total issues

19

46

Percentage found

73.68%

78.26%

Difference in percentage

4.58%

Confidence level

31%

The test for statistical significance between two rates (in this case, catching A out of a total of B problems versus catching C out of a total of D problems) is performed as follows:

Let

P1 = A/B

P2 = C/D

P* = (A+C)/(B+D)

Then the test statistic (Z) will be equal to Numerator/Denominator, where

Numerator = the absolute value of (P1 � P2) and

Denominator = the square root of [(P*) * (1-P*) * ((1/B) + (1/D))]

And the confidence interval will equal the area under the standard normal curve at Z, using a two-tailed test.