Assumptions of Multiple Linear Regressions

LINEAR RELATIONSHIP between dependent variable y and EACH explanatory variable

Residuals NORMALLY distributed

Residuals should have CONSTANT VARIATION against range of fitted values for y

Checking Assumptions
Assumption 1
Plot dependent variable against each explanatory variable

Assumption 2
Plot histogram and normal probability plot of residuals
Assumption 3
Plot residuals against each explanatory variable.
NO pattern should be seen
Variability should be constant across range

Assumption 1 plot dependent variable with each explanatory variable.

A. Age vs systolic blood pressure

B. Weight vs SBP

C. height vs SBP
D. DBP vs SBP

E. Serum cholesterol level vs SBP

Assumption 2
Residuals NORMALLY distributed.
Assumption 3

Residuals should have CONSTANT VARIATION against range of fitted values for y.

N.B.
 Normal p-p plot of regression standardized residual and Histogram used for checking Normality
 Scatter plot of standardized residual and the dependent variable used for checking Spread of
values of dependent variable constant over range of x( independent) values
 Scatter plot of dependent variable Vs each independent variable for checking linearity
 Regression analysis of for possible significant predictors using systolic BP as outcome variable

Table 1: Model Summary

Model R R Square Adjusted Std. Error of the Change Statistics

R Square Estimate
R Square Change F Change df1 df2 Sig. F Change

1 .878a .772 .766 9.579 .772 142.187 9 379a .000

a. Predictors: (Constant), high blood pressure , body weight in pound , race of the subject , serum
colestrol level in mg per 100ml, does respondent smoking status, sex of the subject , diastolic blood
pressure, age of the subject in years , height in inches

 R2 = 77.2%
 77.2% of variation in systolic BP is explained by the model ( difference in dependent variables)
 With many variables in model R2 tends be an overestimate

 Adjusted R2 is a more conservative estimate

 adjusted R2 = 76.6%

Table 2: ANOVAa

Model Sum of Squares df Mean Square F Sig.

Regression 117432.346 9 13048.038 142.187 .000b

1 Residual 34779.484 379 91.766

Total 152211.830 388

a. Dependent Variable: systolic blood pressure

b. Predictors: (Constant), high blood pressure , body weight in pound , race of the subject ,
serum cholesterol level in mg per 100ml, does respondent smoke now?, sex of the subject ,
diastolic blood pressure, age of the subject in years , height in inches
Table 3: Coefficientsa

Model Unstandardized Standardized t Sig. 95.0% Confidence Interval for

Coefficients Coefficients B

Lower Bound Upper Bound

B Std. Error Beta

(Constant) 119.197 12.035 9.904 .000 95.532 142.861

age of the subject in

.238 .032 .227 7.403 .000 .175 .302
years

sex of the subject 5.058 1.300 .124 3.891 .000 2.502 7.614

race of the subject -.211 .866 -.006 -.244 .808 -1.914 1.492

body weight in pound .031 .014 .065 2.253 .025 .004 .058

1 height in inches -.813 .182 -.149 -4.457 .000 -1.171 -.454

diastolic blood
.505 .052 .285 9.698 .000 .402 .607
pressure

does respondent
.022 1.086 .001 .021 .984 -2.113 2.157
smoke now?

serum colestrol level in

-.009 .011 -.021 -.811 .418 -.031 .013
mg per 100ml

high blood pressure 26.465 1.416 .566 18.692 .000 23.681 29.249

a. Dependent Variable: systolic blood pressure

Using the automated modeling approach of backward regression analysis among the possible selected
candidates as independent variable (predictors) for the dependent variable of systolic BP: age, sex, body
weight, height, diastolic blood pressure, high blood pressure are significant.
Table Backward modeling approach

Model Unstandardized Standardize t Sig. 95.0% Confidence

Coefficients d Interval for B
Coefficients

B Std. Error Beta Upper Lower

Bound Bound

4 (Constant) 117.712 11.790 9.984 .000 94.531 140.893

age of the .233 .029 .222 8.013 .000 .176 .290
subject in years
sex of the 5.094 1.295 .125 3.934 .000 2.548 7.641
subject
body weight in .031 .014 .064 2.267 .024 .004 .057
pound
height in inches -.811 .182 -.148 -4.464 .000 -1.168 -.454
diastolic blood .499 .051 .282 9.711 .000 .398 .600
pressure
high blood 26.490 1.409 .567 18.798 .000 23.720 29.261
pressure

The multiple regression backward model of SBP for the predictors will be as follows:

SBP = 117.712 + 0.233(Age) + 5.094(sex) + 0.031(bodyweight) - 0.811(height) + 0.499(DBP) +

26.490(high BP)

Interpretation
Interpretation of unadjusted Coefficients
1. The predicted value of SBP increases by 0.233mmhg for each increase in 1year of age, after
controlling for the other independent predictors (((((sex needs revision)))
2. The predicted value of SBP increases by 0.031mmhg for difference in sex, after controlling for
the other independent predictor
3. The predicted value of SBP increases by 0.811mmhg for each increase in 1cm of height, after
controlling for the other independent predictor
4. The predicted value of SBP increases by 0.499mmhg for each increase in 1mmhg DBP, after
controlling for the other independent predictor
 5. The predicted value of SBP increases by 26.490mmhg for high BP, after controlling for the other
independent predictor.
Summary
The objective of the analysis were to identify factors significantly associated with increased systolic
blood pressure. Accordingly age, sex, body weight, height, diastolic blood pressure, high blood pressure
are significantly associated.

The multiple regression model of SBP is:

SBP = 117.712 + 0.233(Age) + 5.094(sex) + 0.031(bodyweight) - 0.811(height) + 0.499(DBP) +

26.490(high BP)