Table 2.
Hierarchical binary logistic regression models for excessive use of social media
| Model 1 | Model 2 | Model 3 | |||||||
|---|---|---|---|---|---|---|---|---|---|
| B | OR | 95% CI for OR | B | OR | 95% CI for OR | B | OR | 95% CI for OR | |
| Constant | 1.246 | 3.475 | 1.455 | 4.243 | 1.399 | 4.053 | |||
| Gender (girl = ref. +) | −0.578 | 0.561 | (0.491, 0.641)** | −0.628 | 0.533 | (0.466, 0.611)** | −0.629 | 0.533 | (0.465, 0.510)** |
| School (grammar school = ref. +) | |||||||||
| Secondary with graduation | 0.183 | 1.201 | (1.023, 1.409) | 0.126 | 1.134 | (0.963, 1.335) | 0.131 | 1.140 | (0.968, 1.343) |
| Vocational | 0.185 | 1.204 | (1.012, 1.432) | 0.106 | 1.112 | (0.927, 1.335) | 0.116 | 1.123 | (0.934, 1.349) |
| Age (in years) | −0.129 | 0.879 | (0.809, 0.954)** | −0.153 | 0.858 | (0.789, 0.933)** | −0.149 | 0.861 | (0.792, 0.937)** |
| Daily smoking (no = ref. +) | −0.082 | 0.922 | (0.767, 1.108) | −0.078 | 0.925 | (0.768, 1.113) | |||
| Binge drinking (no = ref. +) | 0.522 | 1.686 | (1.460, 1.949)** | 0.521 | 1.684 | (1.459, 1.944)** | |||
| Marijuana use (no = ref. +) | 0.081 | 1.085 | (0.932, 1.263) | 0.086 | 1.090 | (0.936, 1.270) | |||
| Family composition (complete family = ref. +) | |||||||||
| Reconstructed | −0.039 | 0.962 | (0.789, 1.172) | ||||||
| Other | −0.069 | 0.933 | (0.798, 1.092) | ||||||
| Hosmer and Lemeshow testa | 7.700 | 3.090 | 3.596 | ||||||
| Sig. | 0.463 | 0.929 | 0.892 | ||||||
| Nagelkerke R2b | 0.027 | 0.045 | 0.046 | ||||||
Note. Dependent variable: excessive social media use (coded as 1), non-excessive social media use (coded as 0). OR: odds ratio.
Hosmer–Lemeshow statistics indicates a poor fit if the significance value is less than 0.05.
Although the Nagelkerke R2 appears low, Hosmer and Lemeshow (2000, p. 167) declare that “low R2 values in logistic regression are the norm and this presents a problem when reporting their values to an audience accustomed to seeing linear regression values.” They advise against routine publishing of R2 values with results from logistic models. However, they find them helpful in the model building state as a statistic to evaluate competing models, which is also the way we present them in our analysis in Tables 2 and 3.
p < .01.