Skip to main content
. Author manuscript; available in PMC: 2022 Aug 1.
Published in final edited form as: Psychiatry Res. 2021 May 19;302:114020. doi: 10.1016/j.psychres.2021.114020

Table 4.

Multiple Linear Regression for Predictors of Social Media Usage from the Parent Perspective Using LASSO-penalized Variable Selection

Bootstrapped
LASSO Parameter
Estimates
Model Outcome and
Predictor Variables* 		Mean
Estimate		 SD 95% CI Standardized
Estimate		 Adjusted
R2
Social Media Usage (overall sample) 0.302
  Intercept 11.103 1.226 8.715 to 13.266 0
  QIDS-C Total 0.176 0.101 0.021 to 0.391 0.169
  Cyberbullying Score 0.144 0.081 0.041 to 0.321 0.126
  Group (MDD vs. normal controls) 2.551 1.474 0.332 to 5.956 0.223
  Salivary Cortisol 10.221 5.175 2.868 to 22.711 0.096
Social Media Usage (MDD sample) 0.134
  Intercept 16.294 1.839 12.220 to 19.338 0
  Salivary Cortisol 18.166 7.381 7.501 to 34.961 0.266
Social Media Usage (Healthy Control Sample) –
  Intercept 12.921 0.822 11.333 to 14.633 0

Note. The LASSO estimates were based on 10,000 bootstrap samples of the model; Mean Estimate = bootstrap parameter estimate (regression coefficient); SD = standard deviation of the mean parameter estimate; 95% CI for the mean parameter estimate; For the 95% CI that does not contain zero (0), the respective mean parameter estimate is statistically significant at alpha = 0.05 (two-tailed); Standardized Estimate = bootstrap standardized regression coefficient; Adjusted R-squared is the model R-squared based on the LASSO-penalized variable selection; Observed sample: N=60 for the overall sample, n = 30 for the MDD subgroup and n = 30 for the normal control subgroup.

Social media usage was assessed using the Bergen Social Media Addiction Scale (BSMAS), which is 6-item scale that measures risk of social media addiction over the past year. Total score on the BSMAS ranges from 6 to 30, with higher scores representing greater risk of social media addiction.

*

Predictor variables were selected from a pool of 9 potential predictor variables via the LASSO-penalized variable selection method (which performs simultaneous variable selection and parameter estimation) in the context of a multiple linear regression model that was based on 10,000 bootstrap samples.

No significant predictors emerged in the normal control sample.