BuzbyTimeRead the requirements guide

Reading the numbers ยท 2.4

Averages, medians and the week that ruins both

Averages, medians and the week that ruins both. What decides it, what it costs, and what usually goes wrong. For a practical comparison point, see Monitask's work scheduling software.

Time data is not symmetricalThe shape of it

Hours per task, per client, per day: all of these are skewed. A few long items and many short ones, with the long tail doing most of the work in any average.

Which means the mean is usually the wrong summary and it is almost always the one reported, because it is the one a spreadsheet produces by default.

Median and the spread

The median says what a typical case looks like. The spread says how much you can rely on that.

Report both. A median task of two hours with a range from twenty minutes to three days is a different planning problem from a median of two hours with everything between one and four, and the mean is identical in both.

The week that ruins the month

One incident, one crisis, one large delivery. It dominates a monthly average and it tells you nothing about a typical month.

Look at the distribution before drawing a conclusion from any summary. Most surprising monthly figures turn out to be one week.

Small samples

One person's month is a handful of data points. A change of ten per cent between two months is noise for almost any individual measure, and reacting to it teaches people that the numbers are arbitrary.

The practical rule: for an individual, look at quarters. For a team, months. For the business, whatever period the decision covers.

Seasonality

Most businesses have a shape to the year and comparing consecutive months across it produces conclusions about the calendar. Compare with the same period last year as well as with last month, and say which comparison you are making.

Trends rather than points

A single figure invites a reaction. A line over twelve months invites a question, which is the more useful response.

Where a report has to contain a single number, put the twelve month line beside it. It costs nothing and it changes the conversation from what happened to what is happening.

The average utilisation trap

An average across a team conceals the person at ninety per cent and the person at forty. Both are problems, they are opposite problems, and the average reports neither.

Look at the distribution. It is the same amount of work and it answers a question the average cannot.

Percentiles, which are easier than they sound

The ninetieth percentile answers a question managers actually have: what does a bad case look like? It is more useful for planning than an average and it is a single spreadsheet function.

Report the median and the ninetieth together and most planning conversations improve immediately.

Excluding outliers, carefully

Sometimes right and always a decision to be stated. An outlier removed silently is a number quietly reshaped, and the outlier is occasionally the most informative point in the set.

Look at it before removing it. It is frequently a data entry error, and occasionally it is the week that explains the quarter.

Rounding in reports

Reporting to two decimal places implies a precision the underlying entries do not have. Round to the nearest hour at team level. The apparent loss is imaginary and the false confidence removed is real.

Charts that mislead by default

A bar chart of monthly averages hides the distribution completely, and it is what every reporting tool produces first.

Where the distribution matters, show it. A simple scatter of individual durations tells a manager more in five seconds than a year of monthly averages, and it prevents the conclusions this entry is about.

Comparing two teams

Almost never valid. Different work, different mixes, different conventions about what gets recorded. Two teams' figures can differ by half for reasons entirely about recording practice.

Compare each with itself over time, which is available, meaningful and does not start an argument.

Where to start

Take any average currently in a monthly report and plot the underlying distribution once. Most people who do this change what the report contains.

Two numbers worth reporting together, always

The middle and the spread. Every summary in this domain is improved by them and almost every report contains only the first.

A caution about dashboards

A dashboard shows what fits. What fits is single figures, which is precisely the summary this entry argues against, and the constraint is the layout rather than anybody's judgement.

Where the distribution matters, put it in the report even if it will not fit on the screen.

Reporting to somebody who wants one number

Give the median, say it is the median, and put the range beside it in the same sentence. That satisfies the request and prevents the misreading, and it takes no more space than the average would.

A short checklist

Median rather than mean. Spread alongside. Quarters for individuals, months for teams. Compare with the same period last year. Look at the distribution before concluding.

One sentence to carry

The average is the summary a spreadsheet offers and rarely the one the data supports.

The distribution people should see

Their own. Somebody looking at the spread of their own task durations learns something immediately useful about how their week is shaped, and it is the one view of this data that needs no caution attached at all.

Rounding in the record itself

Entries rounded to fifteen minutes are fine and entries rounded to the hour are not, because at the hour the rounding exceeds the duration of most tasks. Pick the increment against the shortest thing worth recording.

Also in reading the numbers