How to read a longitudinal contrast sensitivity graph: a worked example

You have taken the test a few times now, and the app draws you a little line connecting the dots. The most recent dot is lower than the one before it, and your stomach drops. Is your vision getting worse? Before you conclude anything, it is worth learning to read the graph properly — because the single most common mistake is treating the last dot as the answer, when the last dot is mostly noise.

The short version: contrast sensitivity has real measurement noise, so a single dip usually means nothing. What you want to read is the shape of the series over time, and there are only a handful of shapes that matter. This post walks through the noise, the four common patterns, the regression-to-the-mean trap, and the conditions that make a home graph trustworthy at all. By the end you should be able to look at your own dots and know which ones to ignore.

First, the noise

Every measurement wobbles, and contrast sensitivity wobbles more than most. Even a carefully validated clinical instrument is not exact twice in a row. The Pelli-Robson chart — one of the most standardized clinical contrast tests — has a published test-retest repeatability of about 0.15 log units, and the smallest change usually treated as clinically meaningful is about 0.3 log units (Pelli, Robson & Wilkins, 1988; Mäntyjärvi & Laitinen, 2001). That means two honest measurements of the same unchanged eye can differ by 0.15 log units just from normal variation, and you should not get interested until a change reaches roughly double that.

An at-home test adds more sources of wobble on top: screen brightness and gamma, viewing distance, room lighting, whether you are wearing your current correction, how tired you are, how much caffeine is on board. Our piece on why screen settings matter covers the display side, and caffeine, alcohol, and sleep covers the you-side. The practical upshot: on a home graph, be more conservative than the clinical 0.3 log-unit threshold, and treat any single reading as one noisy sample of your true value, not the value itself.

Here is a worked baseline to anchor the intuition. Suppose your true log contrast sensitivity is 1.80 and your test noise is about 0.1 log units per session. Six honest monthly readings of an unchanging eye might come out:

Nothing is changing here. Month 1 to Month 2 is a 0.14 "jump," Month 4 to Month 5 a 0.09 "drop" — and both are pure noise. If you had panicked at Month 5, you would have been reacting to nothing. The average of these six (about 1.79) is a far better estimate of your vision than any single dot.

The four shapes

Once you stop reading individual dots and start reading the series, most longitudinal graphs fall into one of four shapes. Here is what they look like and what each is trying to tell you.

1. Stable scatter. The points bounce around a flat line, none of the moves exceeding the noise band. This is what an unchanging, healthy trend looks like — and it is the most common one. The correct response to stable scatter is to keep your baseline and do nothing. Do not hunt for a trend inside noise; there isn't one.

2. Steady decline. Each session sits a little below the one before, and after several sessions the total drop clears the noise band (more than about 0.3 log units, and ideally sustained across three-plus readings). This is the shape that earns a conversation with your eye doctor. A genuine downward trend across many months, under consistent conditions, is exactly the kind of functional change worth investigating — the reason longitudinal tracking exists in the first place, as we argue in why one test isn't enough.

3. Step change. A flat run, then a sudden drop to a new, lower plateau that also holds flat. The shape itself is a clue: a genuine visual process rarely falls off a cliff and then stops, so a clean step often points to something that happened between the two plateaus. Sometimes it is medical (an event, a new medication). Very often, on a home test, it is a setup change — you switched screens, moved to a brighter room, updated your correction, or started testing at a different time of day. Before treating a step as a vision change, rule out a testing change.

4. Dip then recovery. A temporary drop that returns to baseline. This is the signature of a transient cause: a bad night's sleep, a migraine, fatigue, illness, dry eye on a dry day. It is also the expected shape of recovery from something time-limited — the kind of rebound you might see while healing, discussed in our concussion recovery timeline. A dip that recovers is reassuring precisely because it recovered; the story is in the return, not the dip.

The regression-to-the-mean trap

There is one more effect that fools almost everyone, and it deserves its own section: your first result is a bad baseline.

Two things conspire on the first reading. The first is learning — psychophysical tasks like judging faint patterns genuinely improve with practice over the first few attempts, independent of your eyes. The second is regression to the mean, a statistical fact of any noisy measurement: if your first reading happens to land on a low day, the next reading will, on average, sit closer to your true value simply because extreme readings are unlikely to repeat (Barnett, van der Pols & Dobson, 2005). Both push the same direction, so the second test very often looks "better" than the first — for reasons that have nothing to do with improving vision.

The practical fix: do not trust a baseline until you have three or four sessions. Treat the first reading as a warm-up, and define your personal baseline as the average of several early, consistent sessions. This is also why comparisons to a population "normal" are weaker than comparisons to your own established baseline — your average against yourself, measured well, beats your single dot against a textbook.

Making the graph trustworthy

A longitudinal graph is only worth reading if the sessions are comparable. The rules are boring and they are the whole game:

• Same device. Different screens have different brightness, contrast, and pixel pitch. A graph that spans two laptops is plotting the laptops.
• Same lighting. Room light changes effective screen contrast. Test in similar ambient light each time.
• Same correction. Wear your current glasses or contacts every time. Uncorrected refractive error is the single most common reason contrast looks low, and it swamps everything else.
• Same viewing distance. Follow the test's distance guidance consistently.
• Similar time of day and state. Fatigue, caffeine, and time since waking all nudge the number. Testing at a consistent time reduces that source of scatter.

Get those right and the shapes above mean something. Get them wrong and every graph turns into shape 3, a staircase of setup changes masquerading as vision changes.

Note: a longitudinal contrast sensitivity graph is a screening signal of visual function over time. It cannot diagnose or stage any condition, and its reliability depends entirely on consistent testing conditions. A sustained change is a reason to see a clinician — not a self-diagnosis.

What to do next

• Ignore single dips. Only changes beyond the noise band — sustained across several sessions — are worth attention.
• Read the shape. Stable scatter (do nothing), steady decline (see your doctor), step change (find the event or setup change), dip-then-recovery (a transient cause that resolved).
• Discount your first reading. Build your baseline from three or four consistent early sessions, not the first one.
• Standardize conditions. Same device, lighting, correction, distance, and time of day — every time.

If you want to start a series worth reading, you can take a free contrast sensitivity test, then retake it under the same conditions each time. When a real shape emerges — especially a steady decline — bring the graph to your eye doctor. For the deeper case on tracking over months, see why one test isn't enough, and for how the underlying measurement is estimated, adaptive staircase methods.

References

• Pelli, D. G., Robson, J. G., & Wilkins, A. J. (1988). The design of a new letter chart for measuring contrast sensitivity. Clinical Vision Sciences, 2, 187–199. Source of the Pelli-Robson chart and its test-retest repeatability (about 0.15 log units).
• Mäntyjärvi, M., & Laitinen, T. (2001). Normal values for the Pelli-Robson contrast sensitivity test. Journal of Cataract & Refractive Surgery, 27(2), 261–266. Source of the age-stratified normative values and the ~0.3 log-unit smallest-meaningful-change figure.
• Barnett, A. G., van der Pols, J. C., & Dobson, A. J. (2005). Regression to the mean: what it is and how to deal with it. International Journal of Epidemiology, 34(1), 215–220. Clear treatment of the statistical effect that makes second measurements tend toward the average — why your first test is a poor baseline.
• Elliott, D. B., & Whitaker, D. (1992). Clinical contrast sensitivity chart evaluation. Ophthalmic & Physiological Optics, 12(3), 275–280. Evaluation of clinical contrast charts and their variability, underpinning the "read the trend, not the dot" argument.