Table 6

Sequential comparison of multinomial logistic models for the prediction of physical activity frequency in boys (n = 3 225)

Statistics for individual predictors

Model statistics


Model

Age

Pubic hair stage

Voice change

Pubertal timing

BDS

Migback × pubic hair stage

BDS × age

Correct classification

Pseudo -2

Log-

Likelihood

Wald χ2 (df) corr. for model

Nagelkerke's pseudo R2


0 #

Wald χ2 (df) corrected

36.5%

9431.11

39.09 (13.62)

.018

p-value*

< .001


1

Wald χ2 (df) corrected

68.82 (2.92)

37.6%

9319.74

104.58 (16.06)

.050

p-value*

< .001

< .001


2

Wald χ2 (df) corrected

22.77 (2.95)

4.81 (5.56)

1.55 (5.76)

37.7%

9310.25

104.20 (24.40)

.053

p-value*

< .001

.836

.990

< .001


3

Wald χ2 (df) corrected

34.45 (2.92)

7.58 (5.52)

5.31 (5.75

18.80 (5.54)

38.2%

9282.33

112.61 (27.93)

.060

p-value*

< .001

.068

.284

.007

< .001


4

Wald χ2 (df) corrected

33.06 (2.91)

7.14 (5.51)

5.42 (5.74)

18.70 (5.56)

9.90 (8.44)

38.5%

9267.46

115.69 (33.30)

.065

p-value*

< .001

.084

.299

.008

.022

< .001


5

Wald χ2 (df) corrected

28.01

(2.90)

12.70 (5.67)

6.20 (5.75)

19.01 (5.54)

13.86 (8.13)

20.50 (5.75)

14.73 (8.15)

38.3%

9217.06

126.10 (40.51)

.078

p-value*

< .001

.013

.258

.007

.017

.024

.008

< .001


* Adjustment for multiple tests: Šidák sequential

# model 0 = baseline model including body fat percentage and sociodemographic variables: region, SES, migrant background migback = migrant background

Each row of the table shows the results of one tested model. Left-hand the test statistics for the independent variables are given while right-hand information on model fit is displayed.

The corrected Wald chi-square test tests if an individual independent variable (individual predictors) or all independent variables together (model statistics) significantly contribute to the prediction of the dependent variable; it is corrected for the sampling plan.

Correct classification rate is the proportion of participants for whom the tested model could correctly predict the category of the dependent variable (PA frequency).

Pseudo -2 Log-Likelihood: In logistic regression models are compared due to their -2 log-likelihood; since for complex samples no likelihood ratio test is available the values are only descriptive; better fitting models have smaller values.

Nagelkerke's pseudo R2 is a measure of explained variation in the dependent variable that emulates R2 from linear regression.

Finne et al. International Journal of Behavioral Nutrition and Physical Activity 2011 8:119   doi:10.1186/1479-5868-8-119

Open Data