Skip to content
Open navigation
OpenOpen workspace

Worked exampleBanking and marketing

Did clients subscribe more often after an earlier campaign succeeded?

Subscriptions to a term deposit, by the outcome of the earlier campaign: Prior success: 894 of 1,373 (65.1%). Prior failure: 605 of 4,252 (14.2%). The share was 50.9 percentage points higher for Prior success.

The table

Current campaign subscription by earlier campaign outcome. Percentages use each earlier-outcome column as the base.
Current campaign \ Earlier campaign outcomePrior successPrior failureNo prior outcomeTotal
Subscribed89465.1%60514.2%3,1418.8%4,640
Did not subscribe47934.9%3,64785.8%32,42291.2%36,548
Total1,3734,25235,56341,188

Count / Column %. Each column total is the base for that column’s percentages.

Shaded cells are the two shares the answer compares.

Check the reading before you quote it

Pick the statement this table supports, then open each option to see what the counts say. Every figure here is computed from the table above.

Which comparison shows whether clients subscribed more often after an earlier campaign succeeded?

  1. Compare the number of subscribers under each earlier outcome.

    Not supported. Counts follow group size. No prior outcome has by far the most subscribers, 3,141, because it holds 35,563 records; its subscription share is the lowest of the three, 8.8%.

  2. Compare what share of all subscribers came from each earlier outcome.

    Not supported. That divides by the 4,640 subscribers, so it says where subscribers came from: 67.7% from No prior outcome and 19.3% from Prior success. It does not say how often each group subscribed.

  3. Compare the subscription share within Prior success and within Prior failure, each out of its own records.

    Supported. 894 of 1,373 (65.1%) against 605 of 4,252 (14.2%), a gap of 50.9 percentage points. Those are the two groups the question names; No prior outcome is reported beside them, not merged into either.

The figures behind the check

  • Prior success: 894 of 1,373 subscribed (65.1%).
  • Prior failure: 605 of 4,252 subscribed (14.2%).
  • No prior outcome: 3,141 of 35,563 subscribed (8.8%); kept outside the two-group comparison.

What to carry forward

Name both groups and both bases when you quote this gap: Prior success against Prior failure. Comparing Prior success with all other records would fold in No prior outcome and answer a different question.

Reproduce the exact tableOpens the calculator with this table’s counts, labels and category order, so each figure can be checked.

Use this result

  1. Reproduce it

    Open these exact counts in the chi-square calculator. It loads the same table, with the same category order, and computes the percentages, χ² and Cramér’s V in your browser. Change a cell to see how the result moves.

  2. Cite it

    Copy this reference. The data credit and licence below travel with it.

    Did clients subscribe more often after an earlier campaign succeeded? — Crosstabs. https://www.crosstabs.com/examples/bk01-how-did-term-deposit-subscription-vary-with-the-earlier-campaign-outcome

    Source

    Scope
    Selected telephone marketing campaign records from one Portuguese bank.
    Period
    May 2008 – November 2010
    One record is
    one client campaign record (repeated contacts can contribute to the same record)

    Data: Sérgio Moro, Paulo Rita and Paulo Cortez (2014). Bank Marketing. UCI Machine Learning Repository. DOI 10.24432/C5K306.. Licensed under CC BY 4.0. Crosstabs selected, recoded and aggregated the records into this table.

    Download these counts: CSVJSON

  3. Run it on your file

    Open a CSV, Excel or SPSS file to build the same kind of table. The file is read in this browser tab: nothing is uploaded and no account is needed.

    Run a crosstab on your own data

Why ask this

Marketing teams decide whom to call again. If clients who said yes before say yes again far more often, that changes who goes on the list, but only if the comparison is read the right way round.

How to read it

Put the earlier outcome in the columns and divide current subscriptions by each column total. Keep the third column in view: 35,563 records had no earlier outcome at all, far more than the two columns in the headline comparison combined.

Column percentages answer “within each column, how often does each row occur?”. Switch to row percentages and the question turns around: “of the records in this row, how were they split across the columns?”. Both are correct; they answer different questions, so say which one you used.

What it means

Subscription was much more common among records whose earlier campaign succeeded. That is an association in the bank’s own records; it does not show that a previous success causes a new one.

Cramér’s V = 0.320. It summarizes the whole table, all 2 × 3 categories, on a scale from 0 (no association) to 1. It does not say which cells differ, and it is not a measure of impact.

Comparing only Prior success and Prior failure, and only Subscribed against every other row, Cramér’s V = 0.494. That number describes this one contrast, not the whole table.

Limits

Selected campaign records from one bank. The bank chose whom to contact and how often, so the comparison cannot say what would happen if everyone were called again.

Keep exploring