Nonlinear relationships
If the data in a scatterplot show curvature, it may be possible to linearise the data by transforming either the explanatory variable or the response. Logarithmic transformations are often effective for linearising relationships.
A secondary goal is to make the response variability equal at all values of the explanatory variable. A transformation of the response often fixes both problems.
Repeat with a few different scatterplots.