Actions

Work Header

Killings, Happiness, and Reasons He Stares at my Hips

Chapter 4: Model Specification

Notes:

(See the end of the chapter for notes.)

Chapter Text

Table 3 presents the results of a multiple linear regression model examining how traits of the Ripper, manipulation, and content are associated with times of staring.

 

Table 3. Regression of staring times on Ripper traits, manipulation, and content.

 

Coefficient

SE

Sig.

Statistics VIF

Prekill

-.260*

.119

.036

2.514

Party

.878*

.370

.024

1.683

Wordmanipulation

.150

.136

.281

1.397

Tear

.450

.308

.154

1.536

Confession

.021

.452

.963

2.936

Alcohol

.279*

.122

.028

1.161

Happiness

.682*

.197

.002

2.494

Intercept

-.498

 

 

 

R square

.606

 

 

 

Dependent variable: Staring Times

*P < 0.05.

 

An analysis of outlier diagnostics identified one potentially influential observation (Week 4), characterized by a studentized deleted residual of 2.839, a standardized DFBETA for wordmanipulation of 1.358, and a standardized DFBETA for tear of 1.191. Removing this observation slightly increases the model’s explanatory power (R² increases from .606 to .670). The associations for alcohol, happiness, and party become stronger, whereas the predictor prekill becomes non-significant (t-value shifts from 0.036 to 0.157). Despite its influence, this observation is retained to preserve the overall explanatory strength of the model.

The mean value of unstandardized residuals is essentially zero, indicating a well-specified linear regression model. The error term has a slight positive skew (0.489), suggesting no strong evidence of problematic asymmetry. The kurtosis is -0.203, slightly flatter than normal, but it is well within the limits. The results of normality tests are both non-significant (>0.05), meaning they do not significantly deviate from a normal distribution, thus supporting the assumption of normality and the validity of the regression model.

As shown in table 3, variance inflation factor (VIF) scores for all predictors are below 4, suggesting that multicollinearity is not a concern in the regression model. Pearson correlations further confirmed low collinearity among predictors, with the strongest correlation being between happiness and prekill (r = 0.686). Multicollinearity is not a problem in the regression model.

Notes:

I tested interaction/mediation effects, none of them is significant. Heteroscedasticity isn't really a concern, so I'll move on to regression analysis. Leave comments if you detect any problem/error in analysis, sometimes I get confused when creating tables.