ebd1 lecture 8 2010
DESCRIPTION
TRANSCRIPT
General Studies 1
(Community dentistry 1)
Lecture 8
Correlation & RegressionCorrelation & Regression
Dr. Nizam AbdullaDr. Nizam Abdulla
© The University of Adelaide, School of Dentistry
What is Correlation?
• Correlation is a measure of association between two continuous variables
• Eg: To measure/describe relationship between 2 numerical variables (cholestrol and mean blood pressure)– Are the two variables associated?– In which direction does the relationship ~
positive or negative– What is the strength of the association
© The University of Adelaide, School of Dentistry
Direction of correlation
• Positive correlation– If X increases Y will increases
• Negative correlation– If X increases Y will decreases
© The University of Adelaide, School of Dentistry
What is Correlation?
Eg. We want to measure the association between
1. age and knowledge score 2. Cholestrol level and mean blood
pressure
It estimates both the direction and the strength of the relationship between variables.
© The University of Adelaide, School of Dentistry
Is there a Correlation ?
© The University of Adelaide, School of Dentistry
Cholestrol level
10987654
me
an
Bld
Pre
ssu
re180
160
140
120
100
80
60
Cholestrol level
10987654
me
an
Blo
od
Pre
ssu
re
180
160
140
120
100
80
60
Cholestrol level
10987654
me
an
Blo
od
Pre
ssu
re
180
160
140
120
100
80
60
Cholestrol
10987654
me
an
Blo
od
Pre
ssu
re
200
180
160
140
120
100
80 Cholestrol
10987654
me
an
Blo
od
Pre
ssu
re
180
160
140
120
100
80
r = 0.98 r = 0.78
r = 0.62 r = 0.32 r = 0.08
‘r’ = Pearson’s Correlation Coefficient (i.e. a measure of strength of correlation)
© The University of Adelaide, School of Dentistry
Cholestrol level
10987654
me
an
Blo
od
Pre
ssu
re
140
130
120
110
100
90
80
70
Cholestrol
10987654
me
an
Blo
od
Pre
ssu
re
180
160
140
120
100
80
Cholestrol level
10987654
me
an
Bld
Pre
ssu
re
180
160
140
120
100
80
60
r = + 0.98 r = - 0.94
r = 0
Therefore, r can be –1 to +1
r = 0 means “no correlation.
© The University of Adelaide, School of Dentistry
Cholestrol level
10987654
me
an
Bld
Pre
ssu
re
180
160
140
120
100
80
60
Cholestrol level
10987654
me
an
Blo
od
Pre
ssu
re
140
130
120
110
100
90
80
70
Cholestrol
10987654
me
an
Blo
od
Pre
ssu
re
180
160
140
120
100
80
r = + 0.98 r = - 0.94
r = 0
If …… r > 0.75V.good-perfect correlation r > 0.50 – < 0.75Moderate-good correlation r > 0.25 - < 0.50Fair correlation r < 0.25little or no correlation
Colton T: Statistics in Medicine. Little, Brown, 1974.
© The University of Adelaide, School of Dentistry
Eg: We want to measure association between age and knowledge score
What hypothesis ?Ho: ρ = 0 (there is no correlation between age and knowledge score); Ha: ρ ≠ 0 (There is correlation between age and knowledge score)
(‘ρ’) = Population
© The University of Adelaide, School of Dentistry
The correlation (Pearson’s) between age and knowledge score is significantly different from zero (P<.001). In other words, there is significant (linear) correlation between age and knowledge score.
The observed correlation coefficient (r) is -.719, which suggests negative and moderate to good correlation (Colton, 1974).
© The University of Adelaide, School of Dentistry
Correlation..summary
Correlation
Strength of association
Direction of association
Any association? P-value
Scatter plotNegative or
positive
Correlation Coefficient (r)
Correlation
© The University of Adelaide, School of Dentistry
Correlation and Regression• Two most common methods used to describe
the linear relationship between two quantitative (numerical) variables (x and y) are correlation and linear regression
• The correlation coefficient is the statistic that measures the strength of the linear relationship between two variables
• The regression line is a prediction equation that estimates the value of y for any given x
© The University of Adelaide, School of Dentistry
• Application depends upon research question; For example– Is there a linear relationship between the
two variables?• Both correlation and regression
– What is the strength of linear relationship?• Correlation
– What will be the increase in the GPA by increasing the average study hours----Prediction
• We use the regression analysis (y= ßo+ß1x)
Difference between Correlation and Regression
© The University of Adelaide, School of Dentistry
Regression equation
A linear regression line has an equation of the form Y = a + bx, where x is the explanatory variable and Y is the dependent variable.
The slope of the line is b, and a represent the y intercept
0,0 1 2 3x
y
5
10
15
+2.5
+2.5
+1 +1
y = a + (b * x)Dependent = Constant + (slope*Independent)
y = 5 + (2.5 * x)
© The University of Adelaide, School of Dentistry
Cholestrol level
10987654
me
an
Bld
Pre
ssu
re180
160
140
120
100
80
60 Rsq = 0.9238
Simple Linear Regression
Age
Kno
wle
dge
Sco
re
25 26 27 28 29 3024
10
12
14
16
18
20
22
Slope, b isRegression Coefficient
Slope, b isRegression Coefficient
e.g. b = 1.5 means:mean knowledge score will be1.5 points more, when age is a year older.
e.g. b = 1.5 means:mean knowledge score will be1.5 points more, when age is a year older.
© The University of Adelaide, School of Dentistry
Cholestrol level
10987654
me
an
Bld
Pre
ssu
re
180
160
140
120
100
80
60
Cholestrol level
10987654
me
an
Blo
od
Pre
ssu
re
140
130
120
110
100
90
80
70
Cholestrol
10987654
me
an
Blo
od
Pre
ssu
re
180
160
140
120
100
80
b = + 15 b = - 15
b = 0
Therefore, b can be –α , 0, +α
b = 0 means “no relationship. b = + means “upward trend” b = - “downward”
© The University of Adelaide, School of Dentistry
Regression (cont.)
If we fit a straight line to our height and weight data we end up with the following equation:
height = 136.898 + 0.519 weight
This straight line will allow us to predict any person's height from a knowledge of that person's weight.Example: We can predict the height of EBD1 students who weigh 55kg: The prediction is height = 136.898 + (0.519 x 55kg) = 165.44cm
The most interesting parameter in a linear model is usually the slope. If the slope is zero, the line is flat, so there's no relationship between the variables.
In this example, the slope is about 0.519 cm per kg (an increase in height of 0.519cm for each kg increase in weight).
© The University of Adelaide, School of Dentistry
Regression (cont.)
Simple linear regression and multiple linear regression are appropriate when the dependent variable is continuous.
Logistic regression is appropriate when the dependent variable is categorical (or dichotomous).
© The University of Adelaide, School of Dentistry
The effectiveness of fissure sealants in preventing caries has been well established in randomised clinical trials. However, there is less consistent evidence of fissure sealant effectiveness from community dental programs. In addition, there are unequivocal findings on the relationship between the effectiveness of fissure sealants and exposure to fluoridated water.
Methods:
The current cohort study examined 4–15 yr-old children across a period of between 6 mths and 3.5 yrs (mean=2 yrs). Oral health data were obtained as part of regular examinations and questionnaire data on residential and water consumption history was provided by parents or guardians.
Results:
A sub-group of 791 people (mean age=10.5 yrs) was selected with one contralateral pair of permanent first molars at baseline where the occlusal surface of one molar had been fissure sealed while the paired surface was diagnosed as sound. The caries incidence of the fissure sealed occlusal surfaces was 5.6% compared to 11.1% for sound surfaces (p<.001), demonstrating a 50% reduction in caries incidence for sealed vs non-sealed surfaces.
An example of a research finding
© The University of Adelaide, School of Dentistry
Children were divided into 3 categories of per cent lifetime exposure to optimally fluoridated water, 0%, 1–99% and 100%. Per cent reduction in caries increment attributable to fissure sealing increased across fluoridated water exposure categories - a 36.2% reduction was found for children with 0% exposure (p>.05), a 44.8% reduction for children with intermediate exposure (p<.01), and an 82.4% reduction for children with 100% lifetime exposure to fluoridated water (p<.001).
Conclusion:
The effectiveness of fissure sealants in community-based programs may be further improved when coupled with increased lifetime exposure to optimally fluoridated water.
© The University of Adelaide, School of Dentistry
Townsend GC. Biostatistics - The foundation of evidence-based dentistry. A self-instructional teaching manual available from the ICC
Bulman JS, Osborn JF (1989) Statistics in dentistry, British Dental Association, London.
Moore DS, McCabe GP (1998) Introduction to the practice of statistics, W.H. Freeman, New York.
Burt BA, Eklund SA (2005) Dentistry, dental practice, and the community, 6th ed. Elsevier Saunders, St. Louis.
Suggested reading
THE END