Factors associated with overweight: are the conclusions influenced by choice of the regression method?

BMC Public Health

Table 4 Significant variables in the models with linear, gamma, quantile, binary and multinomial logistic regression

Variable	Gamma	Linear	Quantile			Binary logistic	Multinomial logistic
Variable	Gamma	Linear	0.25	0.50	0.75	Binary logistic	Overweight	Obesity
Age	+	+	+	+	+	+	+	+
Marital status^a
With partner	+	+	+	+	+	+	[+]	+
Race/colour^b
Brown	-	-	[-]	-	-	-	-	-
White	--	--	--	--	--	--	--	--
Mother’s schooling^c
9–11 years	[+]	[+]	[-]	0	[+]	[+]	[+]	[+]
12 or more years	++	++	+	0	++	++	[++]	++
Domestic overload^d
High	+	+	+	+	+	+	+	+
Years worked at night	+	+	0	0	+	0	0	0
Self-rated health^e
Poor	+	+	+	+	+	+	+	+
Consumption of fried food^f
1–3×/month	+	+	+++	+	+	++	++	+
1–3×/week	++	++	++	++	++	+	[+]	+++
4–6×/week	+++	+++	++++	++++	+++	++++	++++	++++
Daily	++++	++++	+	+++	++++	+++	+++	++
Physical inactivity^g
Yes	+	+	+	+	+	+	+	+
BMI at age 20 years	+	+	+	+	+	+	+	+

Reference categories: ^a “Without partner”; ^b “Black”; ^c “0–8 years”; ^d “Low”; ^e “Good”; ^f “Never or less than 1×/month”; ^g “No”
The symbols “+” and “-” denote the direction of the association (“+” direct association, and “-” inverse association). The quantity of these symbols indicate the strength of the association (i.e., “++” indicate a stronger association than “+”, and “+++” indicate a stronger association than “++”). The symbol “0” indicate non-significant variables, and the symbol “[]” indicate only non-significant categories

ISSN: 1471-2458