Manual Calculations: First, summarize these data graphically in form of a scatter-plot. Do these data points group reasonable well along a sloped line? All Lines are defined by the general form Y = a + bX, where a is its intercept (the value of Y if X is zero) and b represents the slope (the rate of change in Y with X). Follow these step-by-step instructions to determine the line that best fits your data by obtaining estimates for a and b.
cases [i] |
X |
Y |
Xi- |
SSx = (Xi-)2 |
Yi- |
SSy = (Yi-)2 |
(Xi-) * (Yi-) |
1 |
0 |
-1.02 |
|||||
2 |
0 |
1.42 |
|||||
3 |
0 |
-0.26 |
|||||
4 |
0 |
-0.05 |
|||||
5 |
0 |
1.86 |
|||||
6 |
0 |
0.68 |
|||||
7 |
0 |
0.05 |
|||||
8 |
0 |
0.27 |
|||||
9 |
0 |
2.47 |
|||||
10 |
0 |
-1.39 |
|||||
11 |
1 |
-0.84 |
|||||
12 |
1 |
0.12 |
|||||
13 |
1 |
-0.57 |
|||||
14 |
1 |
-1.09 |
|||||
15 |
1 |
0.21 |
|||||
16 |
1 |
-0.13 |
|||||
17 |
1 |
0.94 |
|||||
18 |
1 |
0.87 |
|||||
19 |
1 |
-0.35 |
|||||
20 |
1 |
-0.40 |
|||||
Σ = |
Indicate the location of the regression line on your scatter plot. To test whether the data points are significantly associated with this line, we partition out the total sums of squares [Σ(Yi-)2] into two components, those explained by the line [b Σ(Xi-)(Yi-)] and those associated with deviations from it [Σ(Yi-)2- b Σ(Xi-)(Yi-)]. Mean squares are obtained by dividing the regression and error sum of squares by their respective degrees of freedom. Calculate your test statistic (F) by dividing the mean squares for the straight line model (MSM) by the mean squares for the error (MSE). Look up the F-ratio in the table of critical values for the F-distribution to obtain significance (p-value = 0.05) for 1 and n-2 degrees of freedom.
Source |
Degrees of Freedom
(df) |
Sum of Squares
(SS) |
Mean Square (MS) |
F-ratio |
Regression |
1 |
b Σ(Xi-)(Yi-) |
SSM/1 |
MSM/MSE |
Error |
n-2 |
Σ(Yi-)2 -
|
SSE/(n-2) |
|
Total |
n-1 |
Σ(Yi-)2 |
|
|
Machine Formula: The initial calculations for a and b can be simplified when these calculation are performed with a computer program:
b = (ΣXY * (ΣX*ΣY/n))
/ ( ΣX2
* ((ΣX)
2 /n))
a = - b *
A summary of linear regression from the American Statistical Association
To double-check your solution, compare with these results