Trend Analysis vs Variance Analysis
Both techniques work on the same numbers and answer different questions. Variance analysis compares what happened with what was planned at a point in time. Trend analysis looks at a series of those points and describes the direction and the rate of travel. Reports that carry only the first are common, and they produce a characteristic kind of surprise: a project that has been within tolerance every month until the month it is not.
Section 5 of the PMBOK® Guide Eighth Edition includes both among the analysis techniques, and the practical skill is reading them against each other.
Two different questions
Variance asks where we are. Spend against plan, progress against schedule, defects against expectation, at this date. It is a snapshot, it is easy to calculate, and it is the natural unit of a monthly report. Its limitation is that a single number carries no information about whether the position is improving, stable or deteriorating.
Trend asks which way we are going. The same measure across six or eight periods, with the shape of the line as the output. It needs history, consistent measurement and a little patience, and it is the only one of the two that supports a forecast. A trend also survives the noise that makes individual variances misleading, which is most of its value.
Reading them together
Small variance, steepening trend. The most under-reacted-to position in project reporting. Everything is within tolerance and the direction says it will not be by the autumn, and because nothing has breached anything, no report requires anybody to say so. This combination is worth escalating on the trend alone.
Large variance, flat trend. Uncomfortable and often stable. Something happened, it is in the numbers, and the project is performing normally since. The correct response is to absorb or reset rather than to launch a recovery effort against a condition that has already stopped.
Noisy data with no trend. Where the measure bounces around, individual variances are meaningless and reacting to them produces expensive churn. The honest reading is that the process has natural variation, and the useful response is either a longer measurement period or a different measure.
Nine coupons and four weeks of drift
An aerospace programme was qualifying a new bonding process, and part of the evidence was a weekly batch of test coupons pulled from production and taken to destruction. The acceptance measure was a first-pass rate against a stated threshold.
The weekly report showed the variance against threshold. For four consecutive weeks the rate was inside tolerance, so the report said within tolerance each time, and nothing else was said about it.
Laid out on the bench in order, the coupons said something different. Each week's specimens were deforming a little further at the same point than the previous week's, and by the fourth week two were cracking through. The measure had moved from comfortably inside the threshold to just inside it, in four even steps, and every individual reading had been acceptable.
Nobody had plotted them. The report had a variance column and no trend, and the test engineer who could see the physical series each week had no route to raise something that was not a breach.
The fifth week breached, which triggered an investigation, which found a curing temperature drifting with an ageing controller. By then about six weeks of production had been bonded under a drifting process and had to be re-inspected, at a cost of roughly eleven weeks of programme time across two work packages.
The change afterwards was a line on a chart. The weekly report kept its variance column and gained a plotted series with the previous eight weeks on it, and a rule that any measure moving consistently in one direction for three periods is raised whether or not it has breached anything. In the following year that rule caught two drifts, both of them cheap to fix at the point they were raised.
What each needs to be trustworthy
A consistent basis. A trend across periods measured differently is an artefact of the measurement. This sounds obvious and is violated constantly by changes in scope, reporting boundaries or definitions partway through a project, and the only defence is stating the basis on the chart itself.
Enough points and an agreed action threshold. Three points make a line and not a trend, and a trend with no stated response is decoration. Agree in advance what movement triggers what action, and the analysis becomes a control rather than a commentary.
For a PMP® candidate, the practical reading is that one technique describes a position and the other a direction, so a scenario where everything is within tolerance and the team is uneasy is asking for the trend. A response that reports compliance has answered only half. Situations where every individual measurement is acceptable and the series is not are the kind a structured PMP exam preparation course keeps in play.
Take the two measures your project reports most often and plot the last eight periods of each. It takes ten minutes. If either line is moving steadily in a direction nobody has commented on, you have found something that your reporting format is currently designed to hide.
Interested in going further?
Reports built on single-period comparisons will always be surprised by slow movements, and slow movements are the expensive ones. Omega's PMP® Exam Preparation works through project controls as direction as well as position.
Variance and trend techniques are described among the analysis methods in the PMBOK® Guide Eighth Edition.
Ad · Amazon affiliate link.
References
A178: SWOT Analysis in Project Management
A182: Multicriteria Decision Analysis Explained
A183: Cost-Benefit Analysis: Why the Cheapest Option Is Not Always Best
A163: Monte Carlo Analysis: What It Tells You and What It Doesn’t
A180: Facilitation Techniques for Difficult Project Conversations
PMP and PMBOK are registered marks of the Project Management Institute, Inc.
