AP Statistics, Unit 5
Exam words: Regression Analysis
The 31 words a written answer in this unit needs. For each one: what it means, a wording that loses the mark and why, and a sentence that uses it the way a full answer would.
Study these words as a deck
Flashcards, Learn, Test and Match in Cards, with your Still learning marks kept, and the booklet to print on A4 or US Letter. Sign in to open it.
The words
scatterplot
A graph of two quantitative variables with one point for each individual.
Loses the mark: “line graph of the two variables”. A line graph tracks one variable over time; two variables measured on the same individuals call for a scatterplot.
Example: A scatterplot of hours studied (x) against quiz score (y) for 25 students shows a moderately strong positive linear association.
explanatory variable
The variable used to explain or predict another; it goes on the horizontal (x) axis.
Loses the mark: “the explanatory variable causes”. Explanatory means it helps predict the response; cause is shown only by a randomized experiment.
Example: Hours of sunlight is the explanatory variable, so it goes on the x-axis when predicting tomato yield.
response variable
The variable being explained or predicted; it goes on the vertical (y) axis.
Loses the mark: “the response is the effect”. A response is what we measure or predict; calling it an effect claims a cause the data may not support.
Example: Tomato yield in kilograms per plant is the response variable in the garden study.
association
Knowing one variable's value tells you something about the other's likely value.
Loses the mark: “the association proves”. An association alone never shows that one variable causes the other.
Example: There is an association between shoe size and reading level among elementary students, but age drives both.
direction
Whether a scatterplot's points trend upward (positive) or downward (negative) from left to right.
Loses the mark: “weak because it goes down”. Going down is a negative direction; strength is how closely the points follow the form.
Example: In direction, the association between outdoor temperature and heating cost is negative.
positive association
Above-average values of one variable tend to occur with above-average values of the other.
Loses the mark: “good association”. Positive refers to direction, not to whether the relationship is desirable or strong.
Example: Arm span and height have a positive association: taller students tend to have longer arm spans.
negative association
Above-average values of one variable tend to occur with below-average values of the other.
Loses the mark: “bad association”. Negative refers to direction only; it says nothing about strength or quality.
Example: Car age and resale price show a negative association: older cars tend to sell for less.
form
The overall pattern of a scatterplot: linear if straight, nonlinear if curved, plus any clusters.
Loses the mark: “linear because it goes up”. A rising cloud can still be curved; form is about straight or curved, not up or down.
Example: The form of the plot of stopping distance against speed is curved, bending upward at higher speeds.
strength
How closely the points follow the form: strong if tightly packed, weak if widely scattered.
Loses the mark: “strong because it is steep”. Steepness is the slope; strength is how tightly the points follow the pattern.
Example: The association between arm span and height is strong, because the points lie close to a straight line.
correlation coefficient
The unitless r, from -1 to 1, measuring direction and strength of linear association.
Loses the mark: “r = 0 means no relationship”. r near 0 rules out only a linear trend; a strong curved relationship can still exist.
Example: The correlation between hours studied and quiz score is r = 0.71, a fairly strong positive linear association.
linear association
An association whose points follow a straight-line pattern; r measures only this kind.
Loses the mark: “r shows any relationship”. r describes only straight-line patterns; check the scatterplot for curves.
Example: The scatterplot shows a strong linear association, so r = 0.92 is a fair summary.
causation
One variable's change directly producing change in another; only a randomized experiment can show it.
Loses the mark: “correlation proves causation”. A strong r can come from lurking variables; only random assignment supports a causal claim.
Example: Although ice cream sales and drownings are correlated, we cannot claim causation because hot weather drives both.
lurking variable
An unmeasured variable that affects the variables studied and can create or hide an association.
Loses the mark: “another factor”. Name the specific variable and explain how it is linked to both variables in the study.
Example: Daily temperature is a lurking variable that raises both ice cream sales and the number of swimmers.
least-squares regression line
,
The line minimizing the sum of squared residuals; it passes through the point of means.
Loses the mark: “line through the most points”. The least-squares line minimizes squared vertical distances; it may pass through few or no data points.
Example: The least-squares regression line for the garden data is predicted yield = 1.2 + 0.35(hours of sunlight).
predicted value
The response value the regression line gives for a given x, estimating the mean response.
Loses the mark: “the actual value”. y-hat is a prediction from the line; the observed y for an individual usually differs from it.
Example: For a plant with 8 hours of sunlight, the predicted value is 1.2 + 0.35(8) = 4.0 kg of tomatoes.
slope
The predicted change in the response for each one-unit increase in the explanatory variable.
Loses the mark: “will increase by”. Say predicted or on average: the slope gives the predicted change, not a guaranteed change for each individual.
Example: The slope of 0.35 means the predicted tomato yield increases by 0.35 kg for each additional hour of sunlight.
y-intercept
The predicted response when x = 0, meaningful only if x = 0 makes sense.
Loses the mark: “the starting value”. Say the predicted value of the response when x is 0, and check whether x = 0 makes sense.
Example: The y-intercept of 1.2 kg would be the predicted yield for 0 hours of sunlight, which is far outside the data, so it has no useful meaning here.
influential point
A point whose removal would noticeably change the slope, intercept or correlation.
Loses the mark: “every outlier is influential”. A point far from the line but at a typical x may barely move it; influence depends on leverage.
Example: The 30-year-old car is an influential point: removing it changes the slope from -850 to -1,200 dollars per year.
residual
The observed response minus the predicted response, y minus y-hat.
Loses the mark: “predicted minus actual”. A residual is observed minus predicted; reversing the order flips every sign.
Example: The plant that yielded 4.6 kg had a residual of 4.6 - 4.0 = 0.6 kg, so it produced more than the line predicted.
residual plot
Residuals plotted against x or the predicted values to check whether a linear model fits.
Loses the mark: “has a pattern so the line fits”. A curve or fan in the residuals means a line misses part of the relationship; random scatter supports a linear model.
Example: The residual plot shows a clear U-shaped curve, so a linear model is not appropriate for stopping distance.
random scatter
Residuals scattered evenly above and below zero with no curve or fan shape.
Loses the mark: “the residuals are all small”. Small residuals are not enough; the key check is that no curve or fan pattern appears.
Example: Because the residual plot shows random scatter around zero, a linear model for arm span and height is appropriate.
standard deviation of the residuals
The typical prediction error: how far responses fall from the line's predictions, in response units.
Loses the mark: “the average residual”. Least-squares residuals always average to zero; s measures their typical size.
Example: The standard deviation of the residuals is 0.4 kg, so predicted yields are typically off by about 0.4 kg.
r-squared
The proportion of variation in the response explained by the linear relationship with x.
Loses the mark: “of the data is explained”. Say the percent of the variation in the response explained by the linear model, not of the data.
Example: With r-squared = 0.81, about 81% of the variation in quiz scores is explained by the linear relationship with hours studied.
extrapolation
Predicting from a regression line for x-values outside the range of the observed data.
Loses the mark: “reliable because r is high”. A strong fit inside the data range says nothing about values outside it.
Example: Predicting the yield for 20 hours of sunlight is extrapolation, since the data only covered 4 to 11 hours.
population regression line
The true line mu_y = alpha + beta x giving the population's mean response at each x.
Loses the mark: “the population line is y-hat = a + bx”. y-hat = a + bx is the sample line; the population line uses the parameters alpha and beta.
Example: The population regression line describes how mean tree height changes with age for all trees in the forest, not just the 25 we measured.
true slope
The population regression line's slope beta: mean change in response per one-unit increase in x.
Loses the mark: “b is the true slope”. b is the sample slope, a statistic; beta is the fixed population value it estimates.
Example: We test whether the true slope relating tree age to mean height is greater than 0.
sample slope
The slope b of a sample's least-squares line; it varies from sample to sample.
Loses the mark: “another sample would give the same slope”. b varies from sample to sample; its sampling distribution is centered at beta.
Example: The sample slope b = 0.42 meters per year is one estimate; a new random sample of trees would give a different b.
standard error of the slope
An estimate of how much the sample slope varies between samples, read from SE Coef.
Loses the mark: “s is the standard error of the slope”. s measures typical prediction error; the SE of b measures how much the slope varies between samples.
Example: The standard error of the slope is 0.08, so sample slopes typically vary by about 0.08 meters per year from beta.
LINER conditions
Slope-inference checks: Linear, Independent, Normal residuals, Equal SD of residuals, Random sample or assignment.
Loses the mark: “r is close to 1 so the conditions are met”. Each condition needs its own check, using a scatterplot, a residual plot and the study design.
Example: To check the LINER conditions, the residual plot shows no curve and an even spread, and the trees were a random sample.
t interval for the slope
,
A confidence interval for beta: b plus or minus t* times the SE of b.
Loses the mark: “1.96 times the SE of the slope”. Slope intervals use t* with n - 2 degrees of freedom, not a z* value.
Example: A 95% t interval for the slope is 0.42 plus or minus 2.069(0.08), so mean height rises by between 0.25 and 0.59 meters per year of age.
t test for the slope
A test of H0: beta = 0 (no linear relationship), using t = b over its SE.
Loses the mark: “SE divided by b”. The statistic is b divided by its standard error, not the other way around.
Example: The t test for the slope gives t = 0.42/0.08 = 5.25 with 23 degrees of freedom and a p-value below 0.001.