Examining the Relationship between Heavy Alcohol Use and Assaults: With Adjustment for the Effects of Unmeasured Confounders

Table 3

Comparison of estimates derived from standard univariate models and estimates derived from univariate models with proxy outcome method alone to account for all confounding effects.

	Violent behaviour			Driving without seatbelt			Driving while under the influence of alcohol
	Odds ratio	95% CI		Odds ratio	95% CI		Odds ratio	95% CI^∧
Estimates from univariate models
Days of having 5+ drinks last month
0	1.00			1.00			1.00
1	1.99^*	1.57	2.52	1.57^*	1.25	1.96	2.94^*	2.68	3.21
2	2.73^*	2.09	3.56	2.14^*	1.64	2.79	4.02^*	3.64	4.44
3 to 4	3.17^*	2.57	3.91	2.68^*	2.14	3.35	6.01^*	5.46	6.62
5 to 8	4.62^*	3.75	5.70	2.64^*	2.08	3.36	8.31^*	7.51	9.19
9 to 30	6.53^*	5.38	7.94	3.55^*	2.87	4.39	10.20^*	9.17	11.34
Estimates from univariate models with proxy outcome method
Days of having 5+ drinks last month
0	1.00						1.00
1	1.27	1.00	1.61				1.88^*	1.71	2.05
2	1.27	0.98	1.66	Estimates not applicable for proxy outcome			1.88^*	1.70	2.07
3 to 4	1.18	0.96	1.46	(will always be equal to 1)			2.25^*	2.04	2.47
5 to 8	1.75^*	1.42	2.15				3.14^*	2.84	3.47
9 to 30	1.84^*	1.52	2.24				2.87^*	2.58	3.20

95% confidence interval.
.