Some slightly curved data appears to be linear and possesses a high correlation coefficient. However, the curve of the data will reveal itself in the residuals, which would produce a curved pattern.

Unusual features on a scatterplot exert too much influence on the linear regression. Sometimes these features display a pattern that isn't actually there. Sometimes it brings the regression line towards its location.

If an unusual feature is within the interval of the other datapoints, it is only an outlier and does not exert much influence.

We can use mathemetical tools such as logarithms, exponentials, and square roots to re-express a curved model as a linear model.