Step Test Analysis: 4 Threshold Methods, 33 Watt Difference
Step test data analysis: what the raw data sheet must contain, a 7-step dataset with 33 watt method deviation, error propagation and quality check.
The analysis of step-test data turns two columns โ the intensity held during each step and the blood lactate value measured at the end of that step โ into a curve and one or more threshold estimates. In the seven-step data set worked through further below, four established methods yield 151 W, 160 W, 180 W and 184 W from identical rows. This 33-watt spread comes from the analysis, not from the measuring device.
The first practical question is therefore not "What value did I get?" but "Which method produced it, and would the same method also be applied to my next test?"
Three unrelated tests share the name
If you arrived here via a groundwater search: step-drawdown test analysis is a different discipline โ a well is pumped at increasing flow rates and the drawdown is split into aquifer loss and well loss. In electronics manufacturing, "test step analysis" refers to pass/fail and measurement data within a production test sequence. This page addresses the third meaning: an incremental exercise test in which the load rises in fixed steps and a capillary blood sample is taken at the end of each step.
Everything that follows refers to this third case, on the bike or treadmill.
What the raw data sheet must contain
Most analysis problems are recording problems that only surface three weeks later, when a second test can't be reconciled with the first.
| Field | What the analysis does with it | Failure mode if missing | |---|---|---| | Step intensity (W, km/h, pace) | x-axis | Target intensity logged instead of actual intensity โ the curve is plotted against a load nobody actually rode | | Step duration | Determines how far blood lactate has equilibrated by the time of sampling | 3-minute and 5-minute steps get compared as if interchangeable | | Lactate (mmol/l, one decimal place) | y-axis | Rounded to whole numbers, which erases the flat part of the curve | | Sampling time relative to the end of the step | Ensures all points sit on the same basis | Sampling drifts across steps from โ20 s to +40 s; the curve tilts | | Heart rate at end of step | Second axis for later comparisons | No way to check whether two tests were conducted under comparable conditions | | Reason for termination | Determines whether the file is analyzable at all | A test aborted by a mechanical problem is analyzed as if it ended due to exhaustion | | Device, test-strip batches, date | Provenance for comparisons between tests | Two data sets, two test-strip batches, no note explaining a systematic deviation |
The most commonly skipped item is the sampling time. Without it, a later reviewer cannot determine whether step 5 was sampled in the final seconds of exertion or half a minute into recovery โ and there is no way to correct for this after the fact.
One data set, four thresholds
The numbers below are a fictional worked example intended to make the arithmetic visible. Thresholds describe a test result; they are not medical or training advice, and health questions or training prescriptions belong in the hands of a qualified professional.
Cyclist, seven steps of 5 minutes each, 20-watt increments:
| Step | Power (W) | Lactate (mmol/l) | ฮ vs. previous step | |---|---|---|---| | 1 | 100 | 1.1 | โ | | 2 | 120 | 1.3 | +0.2 | | 3 | 140 | 1.6 | +0.3 | | 4 | 160 | 2.3 | +0.7 | | 5 | 180 | 3.6 | +1.3 | | 6 | 200 | 5.6 | +2.0 | | 7 | 220 | 8.2 | +2.6 |
Four common methods, the same seven rows:
| Method | Result | How it gets there | |---|---|---| | Fixed value 2 mmol/l | 151 W | Linear interpolation between 140 W (1.6) and 160 W (2.3): 0.4 of a 0.7 rise โ +11.4 W | | Dmax | 160 W | Largest deviation between the measured points and the chord from the first to the last point | | Modified Dmax | 180 W | Same construction, but the chord starts at the step before the first rise of โฅ 0.4 mmol/l, here at 140 W | | Fixed value 4 mmol/l | 184 W | Interpolation between 180 W (3.6) and 200 W (5.6): 0.4 of a 2.0 rise โ +4 W |
Spread: 33 W, or 1.65 step widths, from data that never changed.
None of the four is wrong. They answer different questions, and two of them (the fixed values) answer a question about an absolute concentration rather than about the shape of your curve. This leads to a reporting rule, not a ranking: write the method name directly next to the value, in the same cell. A threshold without its method is not a comparable quantity, and comparing a Dmax result from March with a 4-mmol/l result from September creates a training story that exists only in the spreadsheet.
A useful shortcut: Dmax is defined as the greatest perpendicular distance from the curve to the chord, but since the chord is a straight line, converting every vertical distance into a perpendicular distance multiplies them all by the same cosine factor. The step that maximizes the vertical distance is the same step that maximizes the perpendicular distance. You can determine Dmax on measured points alone by subtraction; the fitted-curve version differs only in that its maximum can fall between two steps.
Why the last step counts for more than the middle ones
Both Dmax variants anchor their chord at the last data point, which makes them structurally sensitive to when the test was ended. If you remove the 220-W step from the example and recalculate, Dmax stays at 160 W and modified Dmax stays at 180 W. The estimates remain stable because the step with the maximum distance sits three steps inside the curve, well away from the moving endpoint.
This stability is a property of this data set, not of the method. If the steep section of a curve sits right at the end โ a test ended one step after the first sharp rise โ redrawing the chord can shift the estimate by a full step width. A workable planning rule: design the protocol so that at least two steps lie above the visible inflection point. If there is only one, treat the value as provisional and note this in the report.
How far a swing of 0.3 mmol/l reaches
Interpolation transmits measurement noise according to the local slope of the curve, which means the same error costs differently depending on where it occurs.
At the 2-mmol/l crossing point in the example, adjacent steps differ by 0.7 mmol/l over 20 W. Shifting the reading at 160 W up by 0.3 to 2.6 moves the estimate to 148 W; shifting it down by 0.3 to 2.0 places the crossing exactly at 160 W. A single decimal place of variation on one step covers a range of 12 watts.
At the 4-mmol/l crossing point, the local slope is 2.0 mmol/l over 20 W โ 2.9 times steeper. The same ยฑ0.3 range shifts the estimate only from about 181 W to about 186 W. Noise in the flat part of the curve is expensive; noise in the steep part is nearly harmless.
Two things follow from this. First: if you use a low fixed value like 2 mmol/l, a duplicate sample on the surrounding steps brings more precision than additional steps at the top end. Second: reporting 163 W from a protocol with 20-watt increments is arithmetic dressed up as resolution โ round to the nearest 5 W unless your steps are 10 W or finer.
Seven checks before you fit a curve
Run these on the data sheet, not on the chart. A fitted curve conceals exactly the problems you're looking for.
1. Step duration at each step within ยฑ10 s of the protocol. Flag any step ended early. 2. Sampling window identical at every step and every test date โ for example, the last 20 seconds of exertion. Note the window in the file. 3. Sampling hygiene: puncture site wiped dry, first drop discarded. Sweat contamination is a common reason why a single step comes out implausibly high. 4. Monotonicity over the first three steps. A drop of more than 0.3 mmol/l early in the test points to the sample, not to physiology โ verify it rather than smoothing it away. 5. At least two steps above the inflection point, per the anchoring problem described above. 6. Baseline value captured before step 1, plus device, test-strip batch, and expiration date. 7. Reason for termination recorded. A test ended by cramp, a mechanical failure, or a phone call is flagged, not silently analyzed.
What a calculator actually does with your rows
Whether you're searching for an LT1 calculator or an LT1/LT2 calculator โ the tool behind the label runs the same four steps: read and validate rows, fit a curve (usually a third-degree polynomial or a spline), apply one or more threshold definitions to that fit, and then convert the results into pace, power, and heart-rate zones. LT1 and LT2 are labels for the first and second identified change point on the curve โ and different methods define these points differently, which is why two tools can agree on the same data and still arrive at different results.
The question to ask of any tool is therefore narrow: which curve fit, and which method definition? If the output shows only a single number with no method named, it can't be compared to anything โ not even to its own next run. Lactate test online calculator: enter data and read results covers the input side, and Calculate lactate curve online: step-test input and threshold output covers what happens between the rows and the chart. VLamax estimates fall outside this workflow โ they require input data that a standard step test does not provide.
Spreadsheet or dedicated software
A defensible dividing line: one athlete, one method, two tests per year โ a spreadsheet with a documented formula column serves the purpose, and the arithmetic in this article is all you need. Above about four tests per year, or with more than about three athletes, the spreadsheet stops being the analysis and becomes a risk. The failure mode is boringly consistent: hard-wired cell ranges that silently exclude a seventh step, a copied file where the interpolation still points at last season's rows, and no record of which version produced the number in the training plan. Lactate test analysis software: how to choose the right tool lays out what to check before switching.
Frequently asked questions
What does a step test test for? In this context, it maps the relationship between a step-wise increasing load and blood lactate concentration, so that the shape of this relationship can be described and re-measured later. In hydrogeology, the same term describes a well test; in production testing, it describes a test sequence.
Can step-test data analysis be done in Excel? Yes. Fixed-value thresholds need an interpolation formula, and Dmax needs a chord column plus a subtraction and a MAX. Document the formula next to the result so a later reader can trace the number.
How many steps are needed? Enough that at least two lie above the point where the curve visibly steepens โ typically six or more at 20-watt increments for a trained cyclist. Fewer steps not only reduce detail but also shift the chord that two of the four methods depend on.
The file you should leave out of the comparison
Keep the data set where step 2 was ridden 15 watts above target, or where the sampling window drifted, or where the test ended for a reason unrelated to intensity. Save it, flag it, and exclude it from the trend. A file you cannot reproduce is a file you cannot compare โ and comparability is the whole reason for testing twice.