Traffic Is Uneven Across the Week, Exactly as It Always Has Been
← Chapter 168
Capstone 9 · Web Analytics Memo
Plain-language Brief

Traffic Is Uneven Across the Week, Exactly as It Always Has Been

Weekends run about a fifth below weekdays, and this quarter matches last year almost perfectly.

To  Head of Marketing
From  Web Analytics
Re  Quarterly traffic by day of week
Where this comes from
Chapter Chapter 168 · Chi-Square Goodness-of-Fit: Weekday Traffic
Part Part XXVII · Capstone Projects: Statistical Testing
Dataset capstone-website-visits-by-weekday.xlsx
Notebook View the analysis

Recommendation

Bottom line

There is no anomaly here and nothing to investigate. Traffic is not spread evenly across the week, but it never has been: weekends are quieter, as they are for most business sites. Compared against last year's day-of-week pattern, this quarter is statistically indistinguishable from expectation. Plan campaigns around the midweek peak, and treat the weekend dip as a fixed feature rather than a problem to solve.

What we found

Across 3,514 visits, midweek is the busiest stretch (Wednesday highest at 578) and Sunday the quietest at 373. Weekdays run roughly a fifth above weekend days. That pattern is consistent and systematic rather than random fluctuation.

Bar chart of visits by weekday with a dashed uniform-expectation line; weekdays above, weekend below.
Figure 1. Visits by day, with a dashed line showing what a perfectly even week would look like. Every weekday sits above it and both weekend days sit below it.

The question that was actually asked

We were asked whether traffic is spread evenly across the week. Statistically the answer is a clear no. But that question is worth pausing on, because nobody running a website expects an even week. Testing against an expectation no one holds gives a technically correct answer that tells you nothing you can act on.

Bar chart of chi-square statistics: 78.32 against an even spread across the week, 0.41 against last year's own pattern.
Figure 2. The same data against two different expectations. Only the second question was worth asking.

The more useful question

So we asked a second question: does this quarter look like last year? Measured against last year's day-of-week shares, the answer is that it matches almost exactly (p = 0.999, where anything above 0.05 means no detectable difference). The same seven numbers that looked dramatic against a flat line look completely ordinary against precedent.

Two panels showing the same observed traffic against a uniform expectation (rejected) and against last year's pattern (not rejected).
Figure 3. The identical traffic bars compared twice: against a flat week on the left, and against last year's pattern on the right. Only the comparison changed, not the data.

What to take from this

Two statements are both true: traffic is uneven across the week, and traffic is entirely normal. Which one you report depends on what you compare against. For planning purposes the second is the one that matters, and the practical implication is simply to schedule around the midweek peak.

← Back to Chapter 168📓 Companion notebook📊 Download dataset
From Statistics, Data Science and AI: A Visual Handbook by John Fisher. Every statistic, table, and figure in this report is reproduced by the companion notebook.