Business Analytics Quiz: Avoiding Misleading Charts
10 questions · exam conditions
0:00
Avoiding Misleading ChartsQuestion 1 of 10

A finance dashboard compares annual revenue of $98\$98 million last year with $102\$102 million this year. The designer proposes vertical bars whose axis begins at $96\$96 million and ends at $104\$104 million. On that scale, the displayed height above the baseline for this year would be three times the displayed height for last year.

Which revision would most accurately preserve the comparison without visually overstating the change?

Use a zero baseline for the bars and label the approximately 4.1%4.1\% year-over-year increase.
Retain the truncated baseline but add the exact revenue value above each of the two bars.
Begin the axis at $98\$98 million so the earlier bar represents the minimum observed value.
Use a logarithmic axis and describe the difference as a $4\$4 million absolute increase.
← Back to quizzes

Business Analytics Quiz

Business Analytics Quiz: Avoiding Misleading Charts

Practice Avoiding Misleading Charts in Business Analytics 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 Avoiding Misleading Charts, giving you a quick way to practice the rules, question types, and explanations that matter most for Business Analytics.

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

A finance dashboard compares annual revenue of $98\$98 million last year with $102\$102 million this year. The designer proposes vertical bars whose axis begins at $96\$96 million and ends at $104\$104 million. On that scale, the displayed height above the baseline for this year would be three times the displayed height for last year.

Which revision would most accurately preserve the comparison without visually overstating the change?

  1. Use a zero baseline for the bars and label the approximately 4.1%4.1\% year-over-year increase. (correct answer)
  2. Retain the truncated baseline but add the exact revenue value above each of the two bars.
  3. Begin the axis at $98\$98 million so the earlier bar represents the minimum observed value.
  4. Use a logarithmic axis and describe the difference as a $4\$4 million absolute increase.
Explanation: Whenever you see a question about data visualization, ask yourself: does the visual representation accurately reflect the true magnitude of the difference? This is the core issue of truncated axes and misleading charts. Here, revenue grew from $98M\$98M to $102M\$102M — a change of $4M\$4M on a base of $98M\$98M, or roughly 4.1%4.1\%. That's a meaningful but modest gain. The problem is the proposed axis ($96M\$96M to $104M\$104M) makes the bars appear in a 3:13:1 height ratio, wildly exaggerating a real-world 10298≈1.04:1\frac{102}{98} \approx 1.04:1 ratio. The fix must correct the visual distortion, not just patch around it. A is correct because starting the axis at zero makes bar heights proportional to actual values. The bars would be nearly equal in height, which honestly reflects the 4.1%4.1\% difference. Labeling the percentage explicitly adds a second layer of clarity for the audience. B is wrong because adding labels to a distorted chart doesn't fix the distortion — readers still perceive a 3:13:1 visual ratio, and labels are often ignored or overlooked in dashboard contexts. C is wrong because shifting the baseline to $98M\$98M makes the earlier bar disappear entirely (height of zero), which is even more misleading than the original proposal. D is wrong because a logarithmic axis compresses large differences and is inappropriate for comparing two values this close together; it adds complexity without solving the honesty problem. Study tip: On data visualization questions, always check whether the proposed fix resolves the root cause (distorted scale) or merely adds a cosmetic label that leaves the misleading graphic intact.

Question 2

A forecasting model predicts monthly demand for the next year. Its point forecasts rise gradually from 10,00010{,}000 to 11,50011{,}500 units. Model validation indicates that forecast errors increase with the forecast horizon: the central 80%80\% prediction interval is about ±500\pm 500 units next month and ±2,400\pm 2{,}400 units in month twelve.

Which forecast presentation would avoid conveying unjustified precision to inventory planners?

  1. Display only point forecasts rounded to individual units so planners can enter precise replenishment quantities.
  2. Display the point forecasts with prediction bands that widen by horizon and explain what coverage the bands represent. (correct answer)
  3. Display a constant ±500\pm 500-unit band because using one width makes uncertainty comparable across months.
  4. Display only the upper prediction bound because stockouts are more costly than excess inventory in this setting.
Explanation: Whenever a question asks about communicating forecast uncertainty, your job is to match the presentation to the actual structure of the uncertainty — neither hiding it nor distorting it. Here, the model's uncertainty genuinely grows with horizon: ±500\pm 500 units at month one versus ±2,400\pm 2{,}400 at month twelve. The honest, decision-useful presentation is to show widening prediction bands alongside the point forecasts and explain what "80% coverage" means. That is exactly what B does — it gives planners both the central estimate and a calibrated picture of how confident (or uncertain) the model is at each horizon, letting them make smarter inventory decisions without implying false precision. A is wrong because rounding point forecasts to individual units actually increases the false impression of precision. Presenting "10,347 units" with no uncertainty range implies the model is far more accurate than it is. C fails because applying a constant ±500\pm 500-unit band to all twelve months misrepresents the model's known behavior — it understates uncertainty at longer horizons (months 2–12) and falsely suggests the model is equally confident at every point. Planners relying on month-twelve forecasts would be dangerously under-informed. D is wrong because showing only the upper bound discards information (the lower bound and the point forecast) and prevents planners from understanding the full range of outcomes; selectively reporting one tail is not a substitute for communicating the full prediction interval. A useful rule of thumb: the width of your uncertainty display should always mirror the width of your actual uncertainty. When validation shows growing errors by horizon, your bands must grow too.

Question 3

A market report uses circles to represent regional sales. Region X has sales of $40\$40 million, and Region Y has sales of $10\$10 million. The designer sets each circle's radius proportional to sales, so Region X's radius is four times Region Y's radius.

What is the primary problem with this encoding, and which change would correct it?

  1. The areas differ by a factor of 1616; set circle area, rather than radius, proportional to sales. (correct answer)
  2. The circumferences differ by a factor of 1616; set circle diameter proportional to sales instead.
  3. The areas differ by a factor of 44; retain the radii but label both values directly on the chart.
  4. The areas differ by a factor of 1616; use color intensity rather than size to encode the sales difference.
Explanation: Whenever you see a question about encoding quantitative data with circles, your first instinct should be to think about what the human eye actually perceives — and for circles, that's area, not radius or diameter. Here's the core math: area scales with the square of the radius, following A=πr2A = \pi r^2. So when the designer sets Region X's radius four times larger than Region Y's, the area ratio becomes 42=164^2 = 16, not 44. Region X's circle looks 16 times bigger visually, even though sales are only 4 times greater. This creates a dramatic and misleading exaggeration of the difference between $40\$40 million and $10\$10 million. The correct fix is to set area proportional to sales — meaning radius should scale as the square root of sales — so what viewers perceive matches the actual data. That's exactly what A describes, making it the right answer. B is wrong on two counts: circumferences scale linearly with radius (C=2πrC = 2\pi r), so they differ by a factor of 4, not 16 — and switching to diameter doesn't solve the problem because area still scales as radius squared. C correctly identifies the radius-area relationship as the problem but gets the area ratio wrong (it's 16, not 4), and simply labeling values doesn't fix a perceptual distortion. D correctly identifies the factor-of-16 area problem but proposes color intensity as a solution, which introduces a different encoding issue and doesn't address the misleading size comparison. Study tip: Anytime a chart uses circles or bubbles, ask yourself: what does the eye perceive? Perception tracks area, so always verify that area — not radius — is what's mapped to your data variable.

Question 4

An online retailer runs an A/B test. The new checkout has a conversion rate of 10.4%10.4\%, compared with 10.0%10.0\% for the existing checkout. The estimated lift is therefore 0.40.4 percentage points, with a 95%95\% confidence interval from −0.2-0.2 to 1.01.0 percentage points.

Which dashboard presentation would communicate the result without implying more certainty than the test supports?

  1. Show only the two conversion rates because their observed ordering directly identifies the better checkout.
  2. Show a 0.40.4-point lift and label the new checkout successful because most interval values are positive.
  3. Show the estimated lift with its confidence interval and state that the data remain compatible with a small loss or gain. (correct answer)
  4. Show no estimated lift and state that the two checkouts have exactly equal underlying conversion rates.
Explanation: Whenever a question asks how to present statistical results, your job is to match the communication to what the data actually tell you — no more, no less. The key concept here is statistical uncertainty: a confidence interval tells you the range of plausible true values, and a good dashboard respects that range rather than hiding or overstating it. The confidence interval here runs from −0.2-0.2 to +1.0+1.0 percentage points. Because it crosses zero, the test cannot rule out that the new checkout actually hurts conversions slightly. The honest presentation shows the estimated lift (+0.4+0.4 pp) alongside that interval and explicitly acknowledges the result is compatible with both a small loss and a small gain. That is exactly what C does — it gives decision-makers the estimate and its uncertainty in plain language. A is wrong because observed ordering is not reliable inference. Seeing 10.4%>10.0%10.4\% > 10.0\% in sample data does not confirm the new checkout is truly better; sampling variability could produce that gap even if the underlying rates were equal or reversed. B is wrong because "most interval values are positive" is not a valid decision rule. The interval still includes negative values, meaning a loss remains plausible. Declaring success ignores that uncertainty and overstates what the test supports. D is wrong in the opposite direction — overcorrecting into false equivalence. The data do provide an estimate (+0.4+0.4 pp); discarding it and claiming exact equality is just as misleading as overclaiming success. Study tip: When a confidence interval crosses zero, treat the result as inconclusive, not as proof of either equality or a winner — dashboards should always display the interval alongside the point estimate.

Question 5

A retailer reports conversion through online and in-store channels. Last quarter, online accounted for 20%20\% of visits and converted at 10%10\%, while stores accounted for 80%80\% and converted at 25%25\%. This quarter, online accounted for 80%80\% of visits and converted at 12%12\%, while stores accounted for 20%20\% and converted at 30%30\%. A proposed chart shows only the overall conversion rate for each quarter.

Why would the proposed chart support a potentially misleading conclusion?

  1. The overall rate decreases from 22%22\% to 15.6%15.6\% even though conversion improves within both channels. (correct answer)
  2. The overall rate increases from 15.6%15.6\% to 22%22\% because the online channel receives more visits.
  3. The overall rate remains unchanged because channel improvements are exactly offset by the new visit mix.
  4. The overall rate cannot be calculated unless the exact number of visits in each quarter is also provided.
Explanation: Whenever you see a question about overall averages broken down by subgroups, think about Simpson's Paradox — a situation where a trend that appears in aggregated data disappears or reverses when the data is properly segmented. Calculate the overall conversion rate for each quarter by weighting each channel's conversion rate by its share of visits. Last quarter: (0.20×10%)+(0.80×25%)=2%+20%=22%(0.20 \times 10\%) + (0.80 \times 25\%) = 2\% + 20\% = 22\%. This quarter: (0.80×12%)+(0.20×30%)=9.6%+6%=15.6%(0.80 \times 12\%) + (0.20 \times 30\%) = 9.6\% + 6\% = 15.6\%. Notice that online improved from 10%10\% to 12%12\%, and stores improved from 25%25\% to 30%30\% — both channels got better. Yet the overall rate fell from 22%22\% to 15.6%15.6\%. Why? Because the mix shifted heavily toward online, which is the lower-converting channel. A chart showing only 22%→15.6%22\% \rightarrow 15.6\% would misleadingly suggest performance declined, when in fact both channels improved. That makes A correct. B has the direction of the change backwards — the overall rate decreased, not increased, this quarter. C is wrong because the rates do not remain unchanged; the mix shift produces a meaningful drop in the aggregate. D is a distractor appealing to uncertainty — you actually can compute weighted averages from percentages alone once visit shares are given, so exact visit counts are unnecessary. The study tip: whenever overall averages move in the opposite direction from every subgroup, suspect a mix-shift effect. Always ask how the weights changed, not just how the values changed.

Question 6

Over twelve months, a company increased digital advertising expenditure while sales also increased. A dashboard places the two rising series together and titles the display, "Digital advertising drove sales growth." During the same period, the company opened new stores, raised prices, and experienced strong market-wide demand.

Which change would make the dashboard's message most analytically defensible?

  1. Convert both series to percentage changes because standardized units are sufficient to demonstrate causal direction.
  2. Keep the causal title but add the store openings and price increases as annotations on the same time axis.
  3. Retitle the display to describe a positive association and note that the descriptive series do not isolate advertising's causal effect. (correct answer)
  4. Lag the sales series by one month because temporal ordering by itself establishes that advertising caused the increase.
Explanation: Whenever a dashboard pairs two trending variables and assigns a causal headline, your first instinct should be to ask: has causation actually been established, or is this just correlation? That distinction is the heart of this question. The scenario describes a classic confounding problem: digital ad spend and sales both rose, but so did store count, prices, and market-wide demand. Any of those factors — or all of them together — could explain the sales lift. Without controlling for these variables through a proper regression or controlled experiment, you cannot attribute the sales increase to advertising alone. The analytically honest fix is exactly what C proposes: retitle the display to acknowledge a positive association and explicitly note that the visualization doesn't isolate advertising's causal effect. This keeps the data visible while preventing the audience from drawing conclusions the data can't support. A is wrong because converting both series to percentage changes standardizes scale but does nothing to address confounding. Uniform units don't reveal causal direction — that's a logical non-sequitur. B is wrong because annotating confounders on the same chart still leaves the causal title intact. You've acknowledged complexity while simultaneously asserting a causal claim — the contradiction remains. D is wrong because temporal ordering alone is insufficient for causality. Just because advertising spending preceded sales doesn't mean it caused them — market demand also preceded (and accompanied) the sales rise. Lagging the series proves sequence, not causation. Your takeaway: on any analytics question, distinguish association from causation. A visualization can honestly show a trend relationship; it cannot by itself prove a causal mechanism without proper controls.

Question 7

Two software products generated support tickets during the same month. Product A generated 240240 tickets from 12,00012{,}000 active customers, while Product B generated 150150 tickets from 5,0005{,}000 active customers. A manager proposes a bar chart of ticket counts and concludes that Product A has the greater support burden.

Which presentation would provide the most defensible comparison of product reliability while retaining relevant workload information?

  1. Show ticket counts only because the support team must process every ticket regardless of customer base.
  2. Show tickets per active customer, with raw counts included separately to describe total support workload. (correct answer)
  3. Show active-customer counts only because a larger installed base necessarily produces more reliable estimates.
  4. Show each product's percentage of all tickets because ticket share adjusts automatically for exposure differences.
Explanation: Whenever you compare performance across groups of different sizes, raw counts alone are misleading — you need a rate (count divided by exposure) to make a fair comparison. This question tests whether you can identify the right metric while also preserving practical context. Calculate the ticket rate for each product: Product A has 240÷12,000=0.02240 \div 12{,}000 = 0.02 tickets per customer, while Product B has 150÷5,000=0.03150 \div 5{,}000 = 0.03 tickets per customer. Despite generating fewer total tickets, Product B actually has a 50% higher issue rate — meaning it's the less reliable product per user. The manager's bar chart of raw counts inverted the conclusion. Answer B is correct because tickets per active customer surfaces this true reliability difference, while retaining raw counts separately gives the support team the workload information they need to staff and budget appropriately. Both dimensions matter; neither alone tells the full story. Answer A fails because processing every ticket is true operationally, but it doesn't justify using raw counts as a reliability comparison — it conflates workload with product quality. Answer C is wrong because showing only customer-base size tells you nothing about ticket generation; a large installed base doesn't inherently produce more reliable estimates of product reliability without the corresponding ticket rate. Answer D is tempting but flawed: ticket share (A = 61.5%, B = 38.5%) still reflects the size imbalance in the denominator, not the per-customer rate, so it doesn't properly adjust for exposure differences the way a true rate does. As a study rule: whenever group sizes differ, always normalize to a rate before comparing. Raw counts describe volume; rates describe behavior.

Question 8

A customer-revenue distribution groups 100100 customers in a revenue interval from $0\$0 to $10,000\$10{,}000 and 160160 customers in an interval from $10,000\$10{,}000 to $30,000\$30{,}000. A proposed histogram uses bar heights equal to customer counts even though the second interval is twice as wide as the first.

If the purpose is to compare the concentration of customers across revenue ranges, which revision is appropriate?

  1. Use counts as heights because the second interval genuinely contains more customers than the first interval.
  2. Merge both intervals because histograms require all observations to occupy one equal-width revenue range.
  3. Divide both counts by the total of 260260 customers and use the resulting percentages as heights.
  4. Use frequency densities of 1010 and 88 customers per $1,000\$1{,}000, respectively. (correct answer)
Explanation: Whenever you see a histogram with unequal-width intervals, your immediate instinct should be to think about frequency density, not raw counts. A histogram's bar area — not its height — should represent the number (or proportion) of observations. When intervals differ in width, using raw counts as heights visually distorts the data, making wider intervals appear more concentrated than they actually are. The correct fix is D. Frequency density is calculated as count divided by interval width. For the first interval: 10010=10\frac{100}{10} = 10 customers per $1,000\$1{,}000. For the second: 16020=8\frac{160}{20} = 8 customers per $1,000\$1{,}000. These heights now allow fair visual comparison — you can see that customers are actually more concentrated in the lower revenue range, even though the second interval has a higher raw count. That's exactly the kind of insight a well-constructed histogram should reveal. A is wrong because raw counts as heights create a misleading picture when intervals are unequal. The second bar would appear tall simply because it covers a wider range, not because it has higher concentration. B is wrong on two counts: merging intervals discards information, and histograms do not require equal-width intervals — they just require frequency density when widths vary. C is tempting but still incorrect. Converting to percentages (100260≈38.5%\frac{100}{260} \approx 38.5\% and 160260≈61.5%\frac{160}{260} \approx 61.5\%) doesn't account for the width difference — it's the same distortion problem with proportions instead of counts. Your study tip: whenever an exam question involves a histogram, check interval widths first. If they're unequal, frequency density is almost always the right framework.

Question 9

A monthly executive report places advertising expenditure and customer complaints on the same chart. Expenditure uses a left axis spanning $0\$0 to $2\$2 million, while complaints use a right axis spanning 400400 to 700700. The axis limits were selected so the two lines rise at nearly identical angles. Management begins describing the pattern as evidence that advertising creates complaints.

Which redesign most directly reduces both the visual manipulation and the unsupported interpretation?

  1. Keep the dual axes but make both begin at zero, then describe the parallel movement as evidence of causation.
  2. Convert complaints to dollars so both series can share one numerical axis and one common measurement unit.
  3. Report only the correlation coefficient because a numerical association cannot be distorted by chart scaling.
  4. Use aligned separate panels with clearly labeled scales and describe the pattern as an association requiring further analysis. (correct answer)
Explanation: Whenever you see a question involving dual-axis charts, your antenna should go up for two distinct problems: visual distortion (manipulated scales creating false impressions) and causal overclaiming (mistaking correlation for causation). A well-designed redesign must address both issues simultaneously. Option D does exactly that. Separate panels with clearly labeled, independently scaled axes eliminate the visual trick — readers can evaluate each series on its own terms without the artificially matched slopes creating an illusion of tight coupling. The description of the pattern as an "association requiring further analysis" is also precisely correct language: you acknowledge the pattern exists without asserting that advertising causes complaints, which the data alone cannot establish. Option A fails on both counts. Zeroing both axes is an improvement in honesty, but keeping dual axes still invites misleading visual comparisons between unrelated scales. Worse, describing the pattern as evidence of causation doubles down on the core interpretive error — correlation never implies causation without controls, experiments, or theory. Option B sounds clever, but converting complaints to dollars is an arbitrary conversion that manufactures comparability rather than revealing it. You'd be introducing assumptions and distortions just to share an axis, which obscures the data rather than clarifying it. Option C is seductive because correlation coefficients seem objective, but a single number strips away context, trends, and timing — and a coefficient still cannot tell you whether a relationship is causal. Numerical summaries can mislead just as charts can. Study tip: On business analytics questions, always ask two separate questions about any visualization: Is the visual encoding honest? and Does the written interpretation match what the data can actually support? A fix must address both.

Question 10

A logistics team reports that on-time delivery rose steadily from 82%82\% in January to 94%94\% in November and attributes the increase to a process change introduced in January. Historical notes indicate that delivery performance is usually lowest each January and highest in late autumn. The proposed report includes only the January-to-November period from the current year.

Which change would best reduce the risk that the selected time range creates a misleading improvement narrative?

  1. Begin the report in February so the unusually low January observation no longer influences the trend.
  2. Add several prior years at the same monthly frequency and compare current months with corresponding prior-year months. (correct answer)
  3. Replace monthly values with the current year's average so seasonal fluctuations do not distract from the process change.
  4. Retain the selected period but use a trend line because smoothing separates process effects from seasonal effects.
Explanation: Whenever you see a question about trend analysis, watch for seasonality — the recurring, calendar-driven pattern that can masquerade as a real process improvement. That's exactly the trap here. If delivery performance is always low in January and always high in late autumn, then a January-to-November rise tells you very little about whether the process change actually worked. You're likely just observing the normal seasonal cycle. The strongest fix is B: adding prior years at the same monthly frequency and comparing each month to its prior-year counterpart. This year-over-year approach controls for seasonality by asking, "Was November better than previous Novembers, and was February better than previous Februaries?" If the current year consistently outperforms prior years at each comparable point, you have genuine evidence of improvement — not just a seasonal rebound. A makes the problem worse, not better. Dropping the low January value removes evidence rather than contextualizing it, and cherry-picking a start date inflates the apparent trend. C collapses the data into a single annual average, which destroys the monthly resolution needed to detect whether the improvement is real or seasonal. Averaging obscures the very pattern you need to examine. D sounds analytical, but a trend line fitted only to the current year cannot separate seasonal effects from process effects — it simply smooths over both. You need external reference data (prior years) to make that distinction. As a study strategy, whenever a question describes a time series with a known seasonal pattern, immediately ask: Has the analysis controlled for seasonality? Year-over-year comparison is the standard tool for doing so.