Tableau Quiz: Trend Lines And Regression

10 questions · exam conditions

0:00

Trend Lines And RegressionQuestion 1 of 10

A view contains continuous Advertising Spend on Columns and continuous Sales on Rows. The marks suggest that equal percentage increases in Advertising Spend correspond to approximately equal absolute increases in Sales. All Advertising Spend values are positive.

Which Tableau trend model is most consistent with this relationship?

A logarithmic model, because Sales changes with the logarithm of Advertising Spend.

An exponential model, because Sales changes by a constant percentage for each spend increase.

A power model, because both Sales and Advertising Spend must be logarithmically transformed.

A linear model, because equal percentage increases in spend are equal absolute changes on the axis.

← Back to quizzes

Tableau Quiz

Tableau Quiz: Trend Lines And Regression

Practice Trend Lines And Regression in Tableau with focused quiz questions that help you check what you know, review explanations, and build confidence with test-style prompts.

What this quiz covers

This quiz focuses on Trend Lines And Regression, giving you a quick way to practice the rules, question types, and explanations that matter most for Tableau.

How to use this quiz

Try each quiz question before looking at the correct answer. Use the explanations to review missed ideas, then come back to similar questions until the pattern feels familiar.

All questions

Question 1

Which Tableau trend model is most consistent with this relationship?

A logarithmic model, because Sales changes with the logarithm of Advertising Spend. (correct answer)
An exponential model, because Sales changes by a constant percentage for each spend increase.
A power model, because both Sales and Advertising Spend must be logarithmically transformed.
A linear model, because equal percentage increases in spend are equal absolute changes on the axis.

Explanation: When analyzing trend models in Tableau, ask yourself: what kind of relationship exists between the variables based on how the data behaves? The key phrase here is "equal percentage increases in Advertising Spend correspond to approximately equal absolute increases in Sales." This is the mathematical fingerprint of a logarithmic relationship. A logarithmic model takes the form

Sales = a + b \cdot \ln(\text{Spend})

. Because the natural log compresses large values, each successive percentage increase in Spend produces a smaller and smaller absolute change in the input to the log — but crucially, equal percentage multipliers always produce the same additive change in

\ln(\text{Spend})

. So doubling spend from $10 to $20 adds the same

\ln(2)

as doubling from $100 to $200. That constant additive log-input translates to a constant absolute Sales increase. Answer A captures this exactly. B is wrong because an exponential model means Sales grows by a constant percentage for each absolute unit increase in Spend — the opposite of what's described. C is wrong because a power model requires both variables to be log-transformed (

\ln(Y)

vs.

\ln(X)

produces a straight line), meaning equal percentage changes in Spend correspond to equal percentage changes in Sales — not equal absolute ones. D is tempting but incorrect: equal percentage increases in Spend do not become equal absolute changes on a linear axis — that logic is precisely why a log transformation is needed. A handy rule: if "equal % changes in X → equal absolute changes in Y," think logarithmic. If "equal % changes in X → equal % changes in Y," think power model.

Question 2

A scatter plot has continuous Quantity on Columns and continuous Shipping Cost on Rows. Ship Mode is placed on Color, and all ship modes appear in the same pane. The analyst wants one linear trend line for each ship mode rather than one line fitted across all marks.

Which action most directly produces the requested result while retaining Ship Mode on Color?

In the Edit Trend Lines dialog, select the option to allow a separate trend line per color field. (correct answer)
Remove Ship Mode from Color and place it on Detail instead, then add a trend line.
Place Ship Mode on Label so each regression line is annotated with the corresponding ship mode name.
Enable confidence bands in the trend-line options so Tableau displays separate uncertainty regions per ship mode.

Explanation: Whenever you see a question about trend lines in Tableau, focus on what controls how trend lines are segmented — that is, what tells Tableau to fit separate lines for separate groups rather than one line across everything. Tableau's trend line engine is aware of the fields already encoding your marks. When a dimension like Ship Mode is on Color, Tableau knows each color represents a distinct group. In the Edit Trend Lines dialog (accessed via the Analytics pane or the trend line context menu), there is an explicit option to model trend lines "per color field." Selecting this instructs Tableau to run a separate regression for each Ship Mode category while leaving Ship Mode on Color exactly where it is — which is precisely what the analyst wants. That makes A the most direct and correct action. B is tempting but misguided. Moving Ship Mode to Detail would still segment the trend lines by ship mode, but it sacrifices the color encoding, which the question explicitly says must be retained. You'd lose a key visual layer unnecessarily. C confuses annotation with modeling. Placing Ship Mode on Label simply adds text labels to marks — it has no effect on how many trend lines Tableau fits or how those regressions are calculated. D is a red herring. Confidence bands show uncertainty intervals around existing trend lines; enabling them does not create additional lines per group or change the segmentation logic at all. The study tip here: in Tableau, visual encoding fields (Color, Size, Detail) influence trend line segmentation, but you still need to confirm the behavior in the Edit Trend Lines dialog — don't assume Tableau auto-segments just because a dimension is on Color.

Question 3

An analyst adds a linear trend line and displays its $95\%$ confidence bands. A manager claims that approximately $95\%$ of future individual marks should fall inside those bands.

How should the analyst respond?

The claim is correct because confidence bands contain approximately $95\%$ of individual observations by definition.
The claim is correct only when the trend line's p-value is below $0.05$ and $R^2$ exceeds $0.50$ .
The claim is incorrect because the bands represent uncertainty around the estimated trend, not a prediction interval for individual marks. (correct answer)
The claim is incorrect because confidence bands show only the observed minimum and maximum response at each predictor value.

Explanation: Whenever you see a question about trend lines and statistical bands in Tableau, pause to ask yourself: what exactly does this band represent — the line itself, or individual data points? That distinction is everything here. A confidence band around a trend line captures uncertainty about where the true mean of the response lies at each predictor value. In other words, it answers: "How precisely have we estimated the average trend?" If you repeated your study many times, roughly

95\%

of those confidence bands would contain the true regression line. This is fundamentally different from asking where individual future observations will land. What the manager is describing is actually a prediction interval — a wider band that accounts for both the uncertainty in the estimated trend and the natural scatter of individual data points around that trend. Prediction intervals are always wider than confidence bands because individual observations vary more than estimated means do. That's why C is correct: the bands reflect estimation uncertainty, not a forecast envelope for single future marks. Choice A gets the definition backwards — confidence bands do not contain

95\%

of individual observations by construction; that would be a prediction interval. Choice B introduces an arbitrary threshold involving p-value and

R^2

that has no bearing on what type of interval the band represents — those statistics speak to model fit, not interval type. Choice D is simply a fabrication; confidence bands are not min/max envelopes of observed responses. As a study tip, always distinguish three things: the trend line (estimated mean), the confidence band (uncertainty around that mean), and the prediction interval (range for individual future values). Tableau displays confidence bands by default — not prediction intervals.

Question 4

A linear trend line for production cost versus units produced has a nonzero fitted intercept. Because management believes cost must be zero when production is zero, an analyst edits the trend-line options and forces the intercept to zero.

What is the most accurate description of the result?

Tableau only extends the existing line to the origin, leaving its slope and regression statistics unchanged.
Tableau preserves the original slope but suppresses the intercept term only in the displayed equation.
Tableau subtracts the original intercept from every observed cost before fitting the same unconstrained model.
Tableau refits the constrained model through the origin, so the slope and fit statistics can change. (correct answer)

Explanation: Whenever you see a question about trend lines and intercept constraints in Tableau, think about what "fitting" a model actually means — it's not cosmetic. A regression line is computed mathematically to minimize residuals given its constraints, so changing a constraint means rerunning the math entirely. When you force the intercept to zero, Tableau refits the ordinary least squares model under the new constraint

\hat{y} = b_1 x

, rather than simply adjusting the display of the unconstrained model

\hat{y} = b_0 + b_1 x

. Because the slope

b_1

is now calculated without a free intercept term absorbing variation, it will generally shift to a different value. Consequently, fit statistics like

R^2

and the residual standard error also change — sometimes dramatically — making D the correct answer. Choice A is wrong because Tableau doesn't just visually "extend" the line; it recalculates the entire model. Choice B describes a display trick — hiding the intercept in the equation label — but Tableau actually refits the model, not just its annotation. Choice C imagines a preprocessing step where observed values are adjusted before fitting, which is not how constrained regression works; the constraint is applied directly to the regression algorithm, not to the raw data. A useful mental model: think of the intercept as a parameter that absorbs unexplained baseline cost. Remove it, and the slope must compensate, pulling the fitted line in a new direction to minimize error through the origin. On the Tableau exam, any time an option suggests a change is "purely cosmetic" or "leaves statistics unchanged," treat that as a red flag worth scrutinizing carefully.

Question 5

For a linear trend line relating Response Time to Satisfaction, Tableau reports a slope of $-0.82$ , a slope p-value of $0.012$ , and $R^2=0.18$ . The view contains observational data and no randomized intervention.

Which interpretation is most defensible?

Response Time causes Satisfaction to decrease because both the p-value and $R^2$ meet significance thresholds.
A statistically detectable negative association exists, but the model explains only a modest share of the variation. (correct answer)
No meaningful association exists because an $R^2$ below $0.50$ makes the negative slope statistically invalid.
The model explains $82\%$ of Satisfaction variation because the absolute slope is $0.82$ .

Explanation: When interpreting statistical output in Tableau, you need to carefully distinguish between three separate concepts: statistical significance, effect size, and causation. Conflating any of these is a classic exam trap. Here, the slope of

-0.82

tells you the direction and magnitude of the relationship — as Response Time increases, Satisfaction tends to decrease. The p-value of

0.012

is below the conventional

0.05

threshold, meaning this negative association is statistically detectable (unlikely due to chance alone). However,

R^2 = 0.18

means the model explains only

18\%

of the variation in Satisfaction — a modest fit. Answer B correctly synthesizes all three pieces: a real but limited association, with no causal claim. Answer A is tempting but makes a critical logical error — causation cannot be inferred from observational data alone, regardless of p-value or

R^2

. There was no randomized intervention, so confounding variables could explain the pattern. Answer C invents a rule that doesn't exist. There is no threshold requiring

R^2 \geq 0.50

for a slope to be statistically valid; significance and explanatory power are independent properties. Answer D confuses the slope with

R^2

. The slope

-0.82

describes the rate of change per unit, not the proportion of variance explained — that's what

R^2

measures. As a study habit, always ask three separate questions when you see regression output: Is the relationship statistically significant? How much variation does it explain? Can I claim causation? These are never the same question, and Tableau exam items are specifically designed to test whether you keep them separate.

Question 6

A worksheet has Region as a discrete field on Rows and SUM(Sales) as a continuous field on Columns, producing one bar per region. The analyst wants to add a regression trend line directly to this view, but the option is unavailable.

Which change most directly creates a view suitable for a regression trend line?

Move Region to Color and keep SUM(Sales) as the only axis field, consolidating all bars into one pane.
Add a second continuous measure to the other axis, giving each mark both a numeric predictor and a numeric response. (correct answer)
Convert SUM(Sales) to a discrete field so both axes contain fields of matching discrete data types.
Place Region on Detail twice so Tableau has sufficient regional observations to perform a regression calculation.

Explanation: Whenever you see a question about trend lines in Tableau, ask yourself: what does a regression line actually require mathematically? A regression models the relationship between two continuous numeric variables — a predictor on one axis and a response on the other. Without two numeric axes, there's no mathematical relationship for Tableau to fit a line to. In the original view, Region (discrete/categorical) sits on Rows and SUM(Sales) sits on Columns. Because one axis is categorical, there's no continuous X-variable — just labeled bars. Tableau literally cannot compute a regression slope in this context. The fix described in B is exactly right: adding a second continuous measure (say, SUM(Profit) or SUM(Quantity)) to the other axis converts the view into a scatter plot, where every mark has both a numeric X and a numeric Y value. That's the minimum requirement for regression, and Tableau's trend line option becomes available immediately. A is a trap — moving Region to Color still leaves you with only one numeric axis. You'd get a single colored bar, not a scatter plot. The regression option remains unavailable. C goes in the wrong direction entirely; converting SUM(Sales) to discrete removes the only continuous measure you had, making regression even less possible. D misunderstands what drives regression availability — Tableau doesn't require a certain number of observations via the Detail shelf; it requires two continuous axes. As a study tip, remember this rule of thumb: trend lines in Tableau need two continuous axes. If you see a view with even one discrete axis, regression won't work until that axis becomes numeric.

Question 7

An exponential trend model displayed by Tableau can be written as $\widehat{Y}=e^{3.2}e^{0.04X}$ . A user increases $X$ by $10$ units while remaining within the observed data range.

What multiplicative change in predicted $Y$ does the model imply?

Predicted $Y$ is multiplied by about $24.53$ because the intercept contributes $e^{3.2}$ again.
Predicted $Y$ is multiplied by exactly $1.40$ because the exponent rises by $40\%$ .
Predicted $Y$ increases by exactly $0.40$ units because the fitted coefficient is $0.04$ .
Predicted $Y$ is multiplied by about $1.49$ because the ratio is $e^{0.4}$ . (correct answer)

Explanation: Whenever you see an exponential trend model in Tableau, the key insight is that multiplying the input translates into multiplying the output — and the intercept term cancels out when you take a ratio. The model is

\widehat{Y} = e^{3.2} \cdot e^{0.04X}

. To find the multiplicative change when

X

increases by 10, divide the new prediction by the old one:

\frac{e^{3.2} \cdot e^{0.04(X+10)}}{e^{3.2} \cdot e^{0.04X}} = e^{0.04 \times 10} = e^{0.4} \approx 1.492

So predicted

Y

is multiplied by roughly

1.49

, confirming D is correct. A is wrong because it mistakenly adds the intercept's contribution (

e^{3.2}

) a second time. The intercept appears in both numerator and denominator, so it cancels completely — it plays no role in the multiplicative change. B makes a common algebra error: it confuses the exponent rising by

0.4

with the output rising by

40\%

. The correct multiplier is

e^{0.4}

, not

1 + 0.4

. Since

e^{0.4} \approx 1.49

, not

1.40

, this is off. C treats the exponential model like a linear one, claiming

Y

increases by an additive constant. Exponential models produce multiplicative changes, not additive shifts, when

X

changes. Study tip: For any exponential model

\widehat{Y} = e^a \cdot e^{bX}

, a unit increase in

X

multiplies

\widehat{Y}

e^b

. For a change of

\Delta X

, the multiplier is

e^{b \cdot \Delta X}

— the intercept always cancels.

Question 8

A worksheet displays one mark for each customer, with continuous Discount on Columns and continuous Profit on Rows. A linear trend line is added. The analyst then replaces Customer ID on Detail with Customer Segment, leaving all other fields and filters unchanged.

How should the analyst expect the regression results to change?

They remain unchanged because replacing a dimension on Detail does not modify either continuous axis.
They are recalculated from the smaller set of aggregated segment-level marks now present in the view. (correct answer)
They are recalculated from the original customer-level rows, but the displayed line becomes less detailed.
They become unavailable because a trend line requires a unique identifier such as Customer ID on Detail.

Explanation: Whenever you see a question about trend lines in Tableau, the most important thing to ask yourself is: what marks are currently in the view? Tableau's trend line model is fit exclusively against the marks that are rendered — it has no memory of the underlying row-level data once aggregation has occurred. When Customer ID is on Detail, each customer produces its own mark, so the scatter plot contains hundreds of individual points and the regression is fit across all of them. The moment you swap Customer ID for Customer Segment, Tableau re-aggregates: now only a handful of marks exist — one per segment — each representing the summed or averaged Discount and Profit for that segment. The trend line engine re-runs its least-squares calculation on those few aggregated marks, producing a completely different slope, intercept, R², and p-value. That is exactly what B describes, making it correct. A is wrong because it assumes the axis fields alone determine the regression. They determine what is measured, but the number and position of marks — controlled by dimensions on Detail — determine what gets fit. Swapping a dimension on Detail absolutely changes the marks, and therefore the model. C reflects a misunderstanding of how Tableau works: there is no "background" customer-level calculation running in parallel. The displayed marks are the data the trend line sees — there is no hidden granularity being preserved. D is wrong because Tableau does not require a unique identifier for trend lines. Any continuous-by-continuous scatter plot supports a trend line regardless of what dimension is on Detail. Study tip: Always trace the level of detail in the view before reasoning about analytics. The marks visible on screen are the only inputs Tableau's statistical models use.

Question 9

A second-degree polynomial trend line has the fitted equation $\widehat{Y}=100+12X-1.5X^2$ . The observed values of $X$ range from $0$ to $8$ .

Which statement correctly interprets the fitted relationship within the observed range?

Predicted $Y$ increases throughout the range because the linear coefficient $12$ is positive.
Predicted $Y$ decreases throughout the range because the squared coefficient $-1.5$ is negative.
Predicted $Y$ reaches a maximum near $X=4$ and then decreases as $X$ approaches $8$ . (correct answer)
Predicted $Y$ reaches a minimum near $X=4$ and then increases as $X$ approaches $8$ .

Explanation: When you see a second-degree (quadratic) polynomial trend line, your first instinct should be to find the vertex — the point where the curve peaks or bottoms out. The sign of the squared coefficient tells you the shape: negative means the parabola opens downward (a hill), positive means it opens upward (a valley). Here, the equation is

\widehat{Y} = 100 + 12X - 1.5X^2

. Because the coefficient on

X^2

-1.5

(negative), the parabola opens downward, meaning predicted

Y

rises to a maximum and then falls. To find that peak, take the derivative and set it to zero:

\frac{d\widehat{Y}}{dX} = 12 - 3X = 0

, giving

X = 4

. Since

X = 4

falls squarely within the observed range of

0

8

, the model predicts

Y

climbs to a maximum at

X = 4

, then declines — exactly what C describes. A is wrong because it only considers the positive linear term

+12X

and ignores the competing curvature from

-1.5X^2

. At higher values of

X

, the squared term dominates and pulls

Y

back down. B is wrong for the opposite reason — it fixates solely on the negative squared coefficient and concludes

Y

decreases everywhere, ignoring that the linear term drives initial growth. D is wrong because a negative squared coefficient produces a maximum (hill), not a minimum (valley); a minimum would require a positive squared coefficient. Your strategy tip: for any quadratic trend line, immediately check the sign of

X^2

and calculate the vertex using

X = -\frac{b}{2a}

. Then verify the vertex falls within the observed range before drawing conclusions.

Question 10

Tableau fits the logarithmic trend model $\widehat{Y}=24+6\ln(X)$ , where $X$ is monthly website traffic in thousands and $Y$ is monthly support tickets. The analyst compares two otherwise similar months in which traffic doubles from $X=10$ to $X=20$ .

According to the fitted model, what is the approximate change in predicted support tickets?

An increase of about $4.16$ tickets, calculated as $6\ln(2)$ . (correct answer)
An increase of exactly $6$ tickets, because traffic increased by one doubling.
An increase of about $13.86$ tickets, calculated as $6\ln(10)$ .
An increase of exactly $60$ tickets, calculated from the traffic increase of $10$ .

Explanation: When you see a logarithmic trend model, the key insight is that the model responds to ratios of X, not differences. Because

\ln(X)

is involved, doubling traffic doesn't add a fixed number of tickets based on the raw increase — it adds a fixed amount based on the multiplicative factor. To find the change in predicted tickets, calculate

\widehat{Y}

at both values and subtract. At

X=10

\widehat{Y} = 24 + 6\ln(10)

. At

X=20

\widehat{Y} = 24 + 6\ln(20)

. The difference is

6\ln(20) - 6\ln(10) = 6[\ln(20) - \ln(10)] = 6\ln\!\left(\tfrac{20}{10}\right) = 6\ln(2) \approx 6(0.693) \approx 4.16

. That confirms A is correct. Choice B is tempting but wrong — it confuses "one doubling" with a fixed +6 ticket increase. In a logarithmic model, the coefficient 6 scales the log, not the count of doublings directly. The actual increase from one doubling is

6\ln(2) \approx 4.16

, not 6. Choice C makes the error of using

6\ln(10)

— treating the raw traffic difference (

20 - 10 = 10

) as the input to the log rather than the ratio. The difference of the logs simplifies to the log of the ratio, not the log of the difference. Choice D ignores the logarithm entirely, multiplying the coefficient 6 directly by the traffic increase of 10, which treats the model as linear rather than logarithmic. Study tip: For any logarithmic model, a doubling of X always produces the same change in

\widehat{Y}

, equal to

b \cdot \ln(2)

, regardless of the starting value of X — that fixed-per-doubling behavior is the signature of log models.