Multiple regression and Cronbach’s alpha
Call:
lm(formula = NL ~ LR, data = saa)
Residuals:
Min 1Q Median 3Q Max
-2.9902 -0.6875 0.0454 0.6867 2.2630
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.15505 0.06455 -2.40 0.017 *
LR 0.02041 0.00466 4.38 1.6e-05 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.973 on 323 degrees of freedom
Multiple R-squared: 0.0561, Adjusted R-squared: 0.0531
F-statistic: 19.2 on 1 and 323 DF, p-value: 1.61e-05
Call:
lm(formula = NL ~ LR + AgeGroup, data = saa)
Residuals:
Min 1Q Median 3Q Max
-3.0168 -0.6580 0.0846 0.6557 2.2357
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.12920 0.06498 -1.99 0.048 *
LR 0.02770 0.00554 5.00 9.3e-07 ***
AgeGroupOld -0.37194 0.15523 -2.40 0.017 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.966 on 322 degrees of freedom
Multiple R-squared: 0.0726, Adjusted R-squared: 0.0668
F-statistic: 12.6 on 2 and 322 DF, p-value: 5.38e-06
Analysis of Variance Table
Model 1: NL ~ LR
Model 2: NL ~ LR + AgeGroup
Res.Df RSS Df Sum of Sq F Pr(>F)
1 323 306
2 322 300 1 5.36 5.74 0.017 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
saa$LR.z <- (saa$LR - mean(saa$LR)) / sd(saa$LR)
summary(m2 <- lm(NL ~ LR.z + AgeGroup, data=saa)) # LR has larger effect (0.32 per SD; AG: 0.37 in total)
Call:
lm(formula = NL ~ LR.z + AgeGroup, data = saa)
Residuals:
Min 1Q Median 3Q Max
-3.0168 -0.6580 0.0846 0.6557 2.2357
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.0813 0.0634 1.28 0.201
LR.z 0.3214 0.0642 5.00 9.3e-07 ***
AgeGroupOld -0.3719 0.1552 -2.40 0.017 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.966 on 322 degrees of freedom
Multiple R-squared: 0.0726, Adjusted R-squared: 0.0668
F-statistic: 12.6 on 2 and 322 DF, p-value: 5.38e-06
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.081 0.063 1.3 2.0e-01
LR.z 0.321 0.064 5.0 9.3e-07
AgeGroupOld -0.372 0.155 -2.4 1.7e-02
AgeGroup == 'Young'
) with an avg. length of residence (LR.z == 0
) have a predicted nativelikeness of 0.081 Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.081 0.063 1.3 2.0e-01
LR.z 0.321 0.064 5.0 9.3e-07
AgeGroupOld -0.372 0.155 -2.4 1.7e-02
NL = 0.08 + 0.32 * LR.z + -0.37 * AgeGroupOld
AgeGroupOld
equals 1 for the Old
group and 0 for the Young
group Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.081 0.063 1.3 2.0e-01
LR.z 0.321 0.064 5.0 9.3e-07
AgeGroupOld -0.372 0.155 -2.4 1.7e-02
NL = 0.08 + 0.32 * LR.z + -0.37 * AgeGroupOld
LR.z
of 0 and AgeGroup
Young: 0.08 + 0.32 \(\times\) 0 + -0.37 \(\times\) 0 = 0.08 (= Intercept)LR.z
of 0.5 and AgeGroup
Old: 0.08 + 0.32 \(\times\) 0.5 + -0.37 \(\times\) 1 = -0.13
Call:
lm(formula = NL ~ LR.z * AgeGroup, data = saa)
Residuals:
Min 1Q Median 3Q Max
-2.9464 -0.6658 0.0731 0.6751 2.3044
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.1386 0.0684 2.03 0.044 *
LR.z 0.5190 0.1114 4.66 4.7e-06 ***
AgeGroupOld -0.3289 0.1556 -2.11 0.035 *
LR.z:AgeGroupOld -0.2943 0.1360 -2.16 0.031 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.961 on 321 degrees of freedom
Multiple R-squared: 0.0859, Adjusted R-squared: 0.0774
F-statistic: 10.1 on 3 and 321 DF, p-value: 2.37e-06
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.14 0.068 2.0 4.4e-02
LR.z 0.52 0.111 4.7 4.7e-06
AgeGroupOld -0.33 0.156 -2.1 3.5e-02
LR.z:AgeGroupOld -0.29 0.136 -2.2 3.1e-02
NL = 0.14 + 0.52 * LR.z + -0.33 * AGOld + -0.29 * LR.z * AGOld
LR.z
of 0 and AgeGroup
Young: 0.14 + 0.52\(\times\)0 + -0.33\(\times\)0 + -0.29\(\times\)0\(\times\)0 = 0.14LR.z
of 0 and AgeGroup
Old: 0.14 + 0.52\(\times\)0 + -0.33\(\times\)1 + -0.29\(\times\)0\(\times\)1 = -0.19LR.z
of 0.5 and AgeGroup
Old: 0.14 + 0.52\(\times\)0.5 + -0.33\(\times\)1 +-0.29\(\times\)0.5\(\times\)1 = -0.075 Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.14 0.068 2.0 4.4e-02
LR.z 0.52 0.111 4.7 4.7e-06
AgeGroupOld -0.33 0.156 -2.1 3.5e-02
LR.z:AgeGroupOld -0.29 0.136 -2.2 3.1e-02
Young
AgeGroup
, each unit increase of LR
increases NL
by 0.52Old
AgeGroup
, each unit increase of LR
increases NL
by 0.52 + -0.29 = 0.23LR
slope is shifted downwards by 0.23 for the Old
Agegroup
LR
is less beneficial for older people
m4 <- lm(NL ~ AgeGroup * Sex, data=saa) # We drop LR for simplicity (normally you would include it)
summary(m4) # no significant predictors
Call:
lm(formula = NL ~ AgeGroup * Sex, data = saa)
Residuals:
Min 1Q Median 3Q Max
-3.147 -0.648 0.042 0.696 2.142
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.0303 0.0930 -0.33 0.74
AgeGroupOld 0.2448 0.2052 1.19 0.23
SexMale 0.0337 0.1262 0.27 0.79
AgeGroupOld:SexMale -0.3308 0.2718 -1.22 0.22
Residual standard error: 1 on 321 degrees of freedom
Multiple R-squared: 0.00546, Adjusted R-squared: -0.00384
F-statistic: 0.587 on 3 and 321 DF, p-value: 0.624
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.00844 0.0875 0.0965 0.923
AgeGroupOld 0.05616 0.1347 0.4171 0.677
SexMale -0.03760 0.1119 -0.3362 0.737
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.0120 0.0628 -0.191 0.849
AgeGroupOld 0.0549 0.1344 0.408 0.683
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.0200 0.0829 0.241 0.810
SexMale -0.0363 0.1117 -0.325 0.745
A:B
) should be compared to the best model without that term
A
+ B
, if both terms are significantA
, if A
is significant and B
is notB
, if B
is significant and A
is notA
and B
, if neither are significantA
+ B
(if both are not significant), but may not be an improvement over the model without A
and B
.
A
, B
, or their interactionm0a <- lm(NL ~ 1, data=saa) # model without AgeGroup and Sex for comparison
anova(m0a, m4) # interaction is not supported
Analysis of Variance Table
Model 1: NL ~ 1
Model 2: NL ~ AgeGroup * Sex
Res.Df RSS Df Sum of Sq F Pr(>F)
1 324 324
2 321 322 3 1.77 0.59 0.62
AgeGroup
and Sex
is not supportedAgeGroup
is only significant if we also take into account the effect of LR
saa$Age.z <- (saa$Age - mean(saa$Age)) / sd(saa$Age)
summary(m5 <- lm(NL ~ LR.z * Age.z, data=saa)) # Instead of AgeGroup we use numerical Age
Call:
lm(formula = NL ~ LR.z * Age.z, data = saa)
Residuals:
Min 1Q Median 3Q Max
-3.004 -0.670 0.116 0.670 2.237
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 0.0714 0.0623 1.15 0.2520
LR.z 0.4763 0.0923 5.16 4.3e-07 ***
Age.z -0.1752 0.0654 -2.68 0.0078 **
LR.z:Age.z -0.1242 0.0561 -2.21 0.0277 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.959 on 321 degrees of freedom
Multiple R-squared: 0.0882, Adjusted R-squared: 0.0796
F-statistic: 10.3 on 3 and 321 DF, p-value: 1.62e-06
Estimate Std. Error t value Pr(>|t|)
(Intercept) -1.63e-16 0.0535 -3.05e-15 1.00e+00
LR.z 3.32e-01 0.0657 5.06e+00 7.11e-07
Age.z -1.65e-01 0.0657 -2.52e+00 1.23e-02
Analysis of Variance Table
Model 1: NL ~ LR.z + Age.z
Model 2: NL ~ LR.z * Age.z
Res.Df RSS Df Sum of Sq F Pr(>F)
1 322 300
2 321 295 1 4.5 4.89 0.028 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Analysis of Variance Table
Model 1: NL ~ LR.z * AgeGroup
Model 2: NL ~ LR.z * Age.z
Res.Df RSS Df Sum of Sq F Pr(>F)
1 321 296
2 321 295 0 0.724
m5
provides a better fit (RSS lower)
Q3 Q4 Q14 Q15 Q18 Q19 Q28
1 7 2 6 2 2 6 1
2 4 4 3 4 4 4 4
3 3 3 6 6 5 2 6
4 5 4 6 4 3 6 2
5 4 5 6 4 4 2 5
6 4 2 6 2 3 5 2
Some items ( Q3 Q19 ) were negatively correlated with the first principal component and
probably should be reversed.
To do this, run the function again with the 'check.keys=TRUE' option
Reliability analysis
raw_alpha std.alpha G6(smc) average_r S/N ase mean sd median_r
0.41 0.37 0.63 0.078 0.59 0.039 4.2 0.69 0.042
raw_alpha std.alpha G6(smc) average_r S/N alpha se var.r med.r
Q3 0.74 0.69 0.73 0.27 2.18 0.02 0.10 0.36
Q4- 0.74 0.70 0.76 0.28 2.29 0.02 0.09 0.37
Q14 0.84 0.84 0.85 0.46 5.10 0.01 0.02 0.38
Q15- 0.77 0.73 0.80 0.31 2.67 0.01 0.12 0.37
Q18- 0.73 0.69 0.75 0.27 2.18 0.02 0.09 0.35
Q19 0.75 0.69 0.73 0.27 2.21 0.02 0.10 0.35
Q28- 0.71 0.67 0.73 0.26 2.06 0.02 0.08 0.35
R
Thank you for your attention!
https://www.martijnwieling.nl
m.b.wieling@rug.nl