Skip to main content

Table 9 Comparison of prevalence ratios and respective confidence interval estimates (obtained by unadjusted Poisson/Cox regression, Poisson regression with scale factor adjusted by χ2 and deviance, Poisson/Cox regression with robust variances, log-binomial regression and logistic regression) and odds ratio with confidence interval estimate (obtained by logistic regression) with the Mantel-Haenszel prevalence ratio in the analysis of the original and modified data using mother in a paid job as the outcome, father living with the family as the risk factor and social class as confounder (situation3).

From: Alternatives for logistic regression in cross-sectional studies: an empirical comparison of models that directly estimate the prevalence ratio

Original data

Point estimate

95% Confidence interval

 

value

% diff.

lower

upper

width

% diff.

PR Mantel-Haenszel

1.69

--

1.52

1.87

0.35

--

PR Poisson/Cox (unadj)

1.68

-0.3%

1.41

2.00

0.59

68.6%

PR Poisson (χ2)

1.68

-0.3%

1.49

1.90

0.41

17.2%

PR Poisson (deviance)

1.68

-0.3%

1.46

1.94

0.48

35.9%

PR Poisson/Cox (robust)

1.68

-0.3%

1.52

1.86

0.35

-1.5%

PR log-binomial

1.62

-4.0%

1.47

1.78

0.32

-10.4%

OR logistic regression

3.75

122.6%

2.72

5.17

2.45

597.6%

Modified data

Point estimate

 

95% Confidence interval

   
 

value

% diff.

lower

upper

width

% diff.

PR Mantel-Haenszel

1.58

--

1.39

1.79

0.40

--

PR Poisson/Cox (unadj)

1.61

2.0%

1.31

1.97

0.66

63.2%

PR Poisson (χ2)

1.61

2.0%

1.40

1.85

0.45

10.3%

PR Poisson (deviance)

1.61

2.0%

1.37

1.89

0.53

31.0%

PR Poisson/Cox (robust)

1.61

2.0%

1.42

1.83

0.41

2.1%

PR log-binomial

1.71

8.7%

1.53

1.92

0.39

-4.7%

OR logistic regression

2.97

88.1%

2.14

4.11

1.96

385.9%