- Home
- /
- Statistics
- /
- Mistakes
- /
- Relationships & Regression
Common Mistakes in Relationships & Regression
44 common pitfalls across 11 concepts, including 11 recurring stuck points
These are the pitfalls students hit across all 11 relationships & regression concepts, from Two-Way Tables, Relative Frequency and Conditional Relative Frequency onward. Each entry names the concept it comes from, so you can tell a slip from a genuine gap and go straight to the explanation.
Students often know a procedure related to two-way tables but skip the recognition step: Are the same individuals counted by two category questions at once, and does the question want counts or totals from that grid? That leads to a calculation or graph that looks reasonable but answers a different question.
Confusing row vs column percentages
Drawing causation from association
Using the wrong denominator (joint vs marginal total) when calculating conditional percentages
Students often know a procedure related to relative frequency but skip the recognition step: Do you need a raw count turned into a fraction or percent of the total number of observations? That leads to a calculation or graph that looks reasonable but answers a different question.
Comparing raw frequencies across different-sized groups
Forgetting to convert to same format
Rounding too early
Students often know a procedure related to conditional relative frequency but skip the recognition step: Does the question narrow to one row or column of a two-way table first, then ask for a share of that group? That leads to a calculation or graph that looks reasonable but answers a different question.
Using the grand total when a row or column total is needed
Comparing raw counts when the group sizes differ
Confusing conditional relative frequency with conditional probability notation
Students often know a procedure related to correlation but skip the recognition step: Do you need to say whether two measured variables tend to rise together, move oppositely or show no trend, without computing a number? That leads to a calculation or graph that looks reasonable but answers a different question.
Thinking correlation proves causation
Missing confounding variables
Confusing a strong negative correlation with a weak relationship
Students often know a procedure related to correlation vs causation but skip the recognition step: Does a claim say one variable makes another change, when the evidence is only data showing the two go together? That leads to a calculation or graph that looks reasonable but answers a different question.
Assuming correlation proves causation
Missing hidden common causes
Reversed causation
Students often know a procedure related to scatter plot but skip the recognition step: Does each individual give two numbers, to be plotted as one unconnected dot so the shape of the cloud shows the relationship? That leads to a calculation or graph that looks reasonable but answers a different question.
Swapping the independent and dependent variables on the axes — the explanatory variable goes on , the response on
Claiming causation from a scatter plot pattern — correlation does not imply causation; a lurking variable may explain both
Ignoring outliers that could drastically change the correlation or line of best fit
Students often know a procedure related to line of best fit but skip the recognition step: Are you drawing or reading a straight trend line on a scatter plot to summarize the dots and estimate values? That leads to a calculation or graph that looks reasonable but answers a different question.
Forcing line through origin when inappropriate
Using when relationship isn't linear
Ignoring outliers' influence
Students often know a procedure related to linear regression but skip the recognition step: Must you compute the least-squares slope and intercept from the data or summary statistics, then interpret them in context? That leads to a calculation or graph that looks reasonable but answers a different question.
Extrapolating beyond data range
Assuming causation from regression
Ignoring residual patterns
Students often know a procedure related to residuals but skip the recognition step: Do you need how far one actual value sits above or below the value a fitted model predicted for it? That leads to a calculation or graph that looks reasonable but answers a different question.
Ignoring residual plots
Not checking for patterns
Confusing residual with error
Students often know a procedure related to r-squared (coefficient of determination) but skip the recognition step: Do you need the share of the variation in that a fitted model accounts for, from 0 to 1? That leads to a calculation or graph that looks reasonable but answers a different question.
Thinking is always good (overfitting)
Comparing across different datasets
Confusing with correlation
Students often know a procedure related to correlation coefficient but skip the recognition step: Do you need one signed, unitless number between and for how tightly paired data follow a straight line? That leads to a calculation or graph that looks reasonable but answers a different question.
Assuming r measures nonlinear relationships
Confusing correlation with causation
Ignoring outliers that inflate or deflate r