Testing Signifigance and Strength of Relationship in Simple Regression
Although you can do a test of significance on the Pearson correlation (see p. 58) to determine if there is a relationship between the IVs and DVs, this significance testing commonly comes about in a different way. Frequently, the computer printout will list a table of numbers with headings like those in Fig.1175. The coefficients are the intercept (constant), which equals 98.0 and can be identified on the graph. Similarly, the second line of the table is the slope of the regression line, which equals 1.14. The computer also calculates the standard error (SE) of these estimates, using complicated formulas. The t test is the coefficient divided by its SE, with n–2 degrees of freedom (where n is the sample size), and the significance level follows.
Figure 1171 – Figure 6-1: Relationship between blood sugar and skinfold.
Some figures may not display clearly when rendered as a PDF or printed.
Usually, the computer also prints out an “ANOVA table.” But, you say, “I thought we were doing regression, not ANOVA.” We have already drawn the parallel between regression and ANOVA in the previous section. These Sums of Squares end up in the ANOVA table as shown in Fig.1177. If you take the square root of the F ratio, it equals exactly the t value calculated earlier (as it should because it is testing the same relationship).
Figure 1175 – Table 6-1: Output from Regression Analysis
Some figures may not display clearly when rendered as a PDF or printed.
Figure 1177 – Table 6-2: ANOVA Table from Regression Analysis
Some figures may not display clearly when rendered as a PDF or printed.
Finally, the strength of relationship could then be expressed as the ratio of Sum of Squares (SS) (regression) to [SS (regression) + SS (residual)], expressing the proportion of variance accounted for by the IV. In fact, the square root usually is used and is called a Pearson correlation coefficient.
Figure 1448 –
Some figures may not display clearly when rendered as a PDF or printed.
So, in the present example, the correlation is:
Figure 1451 –
Some figures may not display clearly when rendered as a PDF or printed.
We also could have tested significance of the relationship directly by looking up significance levels for different values of the correlation coefficient and different sample sizes. This is, of course, unnecessary at this point.
We can interpret all this graphically by referring back to Fig.1171. In general, the individual data points constitute an ellipse around the fitted line. The correlation coefficient is related to the length and width of the ellipse. A higher correlation is associated with a thinner ellipse and better agreement between actual and predicted values.5343
Content on this page was last changed on March 19, 2009.
© 2002 BC Decker Inc. Show Disclaimer
| 5343. | Norman GR, Streiner DL. PDQ Statistics . 3rd ed. Hamilton, Ontario: BC Decker Inc.; 2003. |
Next Page: Two or More Independent Variables: Multiple Regression »