ISYS3438 Chap.5 Univariate Data and Chart Selection
Univariate Data and Chart Selection
Univariate analysis starts with one variable but branches according to measurement role and question. Categories support frequency, share and ranking, while quantitative values support centre, spread, skew, gaps and outliers. This chapter compares bars, dots, pies, histograms and box plots, then audits Tableau aggregation.
The worked example replaces a single average service-time card with distribution evidence and explains why bin width, median, tail behaviour and record count matter.
What this chapter covers
- 01
nominal and ordinal data
- 02
interval and ratio data
- 03
category counts
- 04
histogram bins
- 05
box plot summaries
- 06
tableau aggregation
Worked example · free
Replace an average with a distribution
- +1Build a histogram and compare defensible bin widths.
- +1Add a box plot or percentile summary to expose spread and tail behaviour.
- +1Report record count and preserve slow cases unless a documented rule excludes them.
Key terms
- Measurement scale
- A measurement scale describes which comparisons and operations are meaningful for a variable.
- Histogram
- A histogram groups quantitative values into intervals to show distribution shape.
- Box plot
- A box plot summarises a quantitative distribution using median, quartiles and potential outliers.
- Aggregation
- Aggregation combines records into a summary such as sum, average or count.
Univariate Data and Chart Selection FAQ
When is a pie chart acceptable?
A pie can show a simple part-to-whole relationship with a few clearly different shares. It is weak for many similar categories or precise ranking, where a sorted bar or dot chart is easier to read.
Why can histogram bins mislead?
Bin width controls which local variation is combined. A narrow setting can create noise, while a wide setting can hide structure. Compare sensible widths and avoid conclusions that depend on one convenient choice.
Why might Tableau classify a postcode incorrectly?
A postcode contains digits but functions as an identifier and category. Convert it to a dimension and check whether import removed leading zeroes before building a view.
Assessment move
Build four views from the same variable: a table, dot or bar view, histogram and box plot where appropriate. Write the question each can answer and the information each hides. Change aggregation and bin settings deliberately, then record which changes alter the conclusion and which only alter appearance. Classify the variable before choosing a display.
Write whether it is nominal, ordinal, interval or ratio and whether the decision concerns frequency, share, rank, centre, spread, shape or unusual cases. State the record grain and population. This short description usually eliminates several unsuitable charts before software opens.
For an identifier encoded as a number, preserve category status and leading zeroes; arithmetic on a branch code or postcode does not create a meaningful measure. For categorical data, compare a frequency table with sorted bars or dots. Preserve an inherent order when categories form stages or ratings, and define the denominator when showing shares.
Use a pie only for a small, clear part-to-whole question where approximate proportions are sufficient. Inspect whether an “other” category hides a meaningful group or whether missing responses are excluded. Counts answer workload questions; percentages answer composition questions. A strong report says which one the decision requires.
For quantitative data, view individual values when feasible, then examine a histogram under several defensible bin widths and a box plot or percentile summary. Record median, quartiles, range and sample size alongside any mean. Investigate outliers before excluding them: they may be entry errors, rare valid cases or evidence of a separate process.
Compare conclusions with and without a documented correction, never with an unexplained deletion. Note when skew makes a single average a poor description of typical experience. Audit aggregation in Tableau by rebuilding one displayed value by hand from a tiny table. Compare SUM, AVG, COUNT and COUNTD, and decide whether rows, customers or subgroup summaries should receive equal weight.
Change filters deliberately to see which records contribute. End the exercise with an evidence sentence, a boundary sentence and a chart-choice justification. The three should align: the visible form supports the claim, the limitation names what one variable cannot explain and the recommendation remains proportionate to that evidence. Finish with a one-variable decision memo.
Define the population and field, state the analytical question, name the selected display and summarise the pattern using centre, spread, frequency or rank as appropriate. Include the valid record count and explain how missing values were handled. Add one sensitivity result, such as an alternative bin width, aggregation or inclusion rule, and say whether it changes the conclusion.
Then identify the second variable or data source needed to explain the pattern, without pretending the univariate view already supplies that explanation. This compact memo is a useful rehearsal for assessment writing because every sentence maps to visible evidence, a documented choice or an explicit boundary. Repeat the analysis on a fresh variable and deliberately choose an unsuitable chart first.
Describe the interpretive cost, then replace it with a defensible form and compare the claims each invites. The contrast strengthens chart-selection reasoning because it connects measurement scale and task to consequences rather than memorising a list of approved chart types.