0
$\begingroup$

According to this site, which is a top google search site, "The higher the coefficient, the higher percentage of points the line passes through when the data points and line are plotted. If the coefficient is 0.80, then 80% of the points should fall within the regression line. " Is this true? Does $100r$ value give us the percentage of points that lie on a linear regression curve.

On another site I see the image I posted below, and it shows very few points that lie on the line of best fit , but the $r$ value which is about 0.95 is relatively high. enter image description here

$\endgroup$
5
  • 2
    $\begingroup$ It sure looks like the line misses a lot more than 8% of the points in your plot! We wouldn’t expect any of the points to be right on the regression line, in fact. $\endgroup$
    – Dave
    Nov 20 '19 at 4:52
  • 2
    $\begingroup$ What a piece of utter nonsense. You should not trust the drivel that Google searches turn up. There's a saying "on the internet nobody knows you're a dog", but apparently there's a few out there giving the game away. $\endgroup$
    – Glen_b
    Nov 20 '19 at 6:33
  • $\begingroup$ @Glen_b-ReinstateMonica quick question, is it called R pearson correlation coefficient or R spearman correlation coefficient. $\endgroup$
    – john
    Nov 21 '19 at 5:35
  • $\begingroup$ The relevant correlation measure for the usual least squares regression line is the Pearson correlation, which is typically denoted r (for a sample correlation at least). In simple regression (one predictor) its also equal to R, which is the correlation between the data and the fitted values. $\endgroup$
    – Glen_b
    Nov 21 '19 at 8:22
  • $\begingroup$ As other users have pointed out, that is not actually the correct interpretation of the $R^2$ coefficient in a linear regression. However, in the particular case of the plot you have included, there are much bigger problems, since the person that created it is trying to fit a straight line through a time-series that is clearly exponential in nature. Population growth over time tends to involve fluctuation in its growth rate, which operates on the existing population in a multiplicative way. Therefore, it is generally best to model it with an auto-regressive model operating on the logarithms $\endgroup$
    – Ben
    Nov 21 '19 at 10:27
2
$\begingroup$

An r of .8 does NOT mean the line goes through 80% of the data points. It means the points are tightly clustered around a line, but it may not go through any of them. The sign of r indicates if the association between the x and y variables is positive or negative, and the square of r gives a measure of what percent of the variability of the data is explained by a linear model.

$\endgroup$
2
  • $\begingroup$ quick question, is it called R pearson correlation coefficient or R spearman correlation coefficien $\endgroup$
    – john
    Nov 21 '19 at 5:36
  • $\begingroup$ I think both Pearson's and Spearman's correlations are denoted with an "r", but i think generally people means Pearson's if they don't specify. $\endgroup$
    – J. Wauters
    Nov 28 '19 at 15:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.