Two devices can move together without giving the same readings

Draft • Research checked 4 October 2026 • AI-assisted DomDNA editorial content. No independent clinical review or publication is claimed. General education, not individual medical advice.

Imagine two measuring tools whose readings rise and fall together, but one consistently reports a larger value. Their correlation could be impressive. Their interchangeability could still be poor.

This distinction matters whenever a summary claims that a new measurement “matches” another method. A relationship between readings is not the same thing as close agreement in the units you actually use.

A simple invented pair of scales

Suppose a hypothetical scale A reads 50, 60 and 70 units for three objects. Scale B reads 60, 70 and 80 units for the same objects. The pattern moves together perfectly, but B is ten units higher every time.

If you care about ranking the objects, both scales preserve that order. If you care about substituting one reading for the other, the systematic difference matters. Neither tool becomes accurate merely because the pattern is tidy.

These are invented units and values. They do not describe a weighing scale, wearable or laboratory instrument. The arithmetic separates two questions that promotional summaries often blend.

Differences need their own examination

Bland and Altman's original paper on agreement and intraclass correlation argues against treating a correlation statistic alone as evidence of interchangeable methods. Their method-comparison work focuses attention on paired differences and their spread.

For a general reader, the practical idea is straightforward. Ask how far apart the two readings are for the same observation, whether one tends to be higher and whether differences become larger at particular measurement levels.

A small average difference is not enough either. Positive and negative differences can cancel while individual pairs remain far apart. The pattern around the average matters when deciding whether substitution is sensible.

The reference method needs a role

A comparison should name the reference and explain why it is appropriate for the question. Comparing two consumer devices does not automatically establish either one's accuracy. They could share an error or use related assumptions.

It also matters whether they measured the same thing at the same time. A daily estimate compared with a short laboratory measurement can mix disagreement between methods with a difference in the underlying observation window.

The headline word “validated” therefore needs a specific object: which measurement, under which conditions, against which comparator and for which intended use? Agreement for one outcome does not prove agreement for all the device's outputs.

An agreement estimate is still an estimate

A 2016 methodological paper on confidence intervals for limits of agreement addresses uncertainty in estimated agreement limits. Those limits do not arrive free of sampling error, and their interpretation involves assumptions.

This article does not supply a calculation for clinical substitution or a universal acceptable error. Acceptability depends on the intended use. An error tolerable for an approximate activity trend may be unacceptable for a decision needing precise measurement.

A study with many repeated readings from a few people also deserves attention to how the analysis handled those observations. A large number of data points is not necessarily a large number of independent participants.

What you can do with this

When someone cites a correlation as proof of accuracy, ask for the paired-difference results and the intended use. You do not need to accuse the comparison of being invalid. You need to know whether it answers the question the claim suggests.

For the imaginary scales, write “same ordering, ten-unit offset” rather than “perfect match”. That sentence is more useful than the correlation alone because it describes the actual relationship in measurement units.

In a research summary, look for plots or tables that show differences across the measurement range, not only a straight line through two sets of readings. If the accessible material reports only correlation, record that agreement was not established there.

Finally, avoid combining readings from different methods into one continuous personal series without noting the change. Even strongly related methods can have an offset or different variability. Preserving the method label gives a later comparison a chance to be honest.

Moving together is useful information. Giving interchangeable answers is a stronger claim, and it needs evidence aimed at that claim.

For an educational starting point, the DomDNA lifestyle quiz records lifestyle answers, not DNA analysis or a clinical diagnosis.