Sequential Sampling & Decide in Fewer Counts

Sequential sampling is a scouting method where you decide whether to keep sampling after every unit you check, rather than counting a fixed number first. Two decision lines are drawn against the number of samples; your running cumulative count walks between them and you stop the instant it crosses the upper 'treat' line or the lower 'don't-treat' line. When the pest level is clearly high or clearly low, the decision comes quickly, saving scouting time.

The tool uses Wald's Sequential Probability Ratio Test. For each pest plan it computes a slope and two intercepts from the lower and upper mean counts that bracket the economic threshold, the count distribution's aggregation parameter k, and your error rates. That gives an upper 'treat' line and a lower 'don't-treat' line. As you enter counts, the cumulative total is compared to both lines each sample; crossing one ends sampling with that decision.

It depends on how far the infestation is from the threshold — clear cases stop very early, borderline cases take longer. Against a representative fixed plan of 30 units, a clearly high or clearly low field can resolve in a handful of samples, saving most of the scouting effort. The tool reports the samples you took at the decision and the number saved versus the fixed plan for your walk.

They are the two parallel stop boundaries. The upper 'treat' line is the slope times the sample number plus the upper intercept; if your cumulative count reaches it, treat and stop. The lower 'don't-treat' line is the slope times the sample number minus the lower intercept; if your cumulative count falls to it, don't treat and stop. Between the lines, the result is still uncertain, so you keep sampling.

The lower 'don't-treat' intercept is negative at the start, so the line sits below zero for the first samples and a count can never fall under it that early — you cannot responsibly conclude 'don't treat' from one or two units. The line rises into positive territory as samples accumulate, at which point a consistently low count can cross it and end sampling with a don't-treat decision.

The action or economic threshold is the pest density at which the cost of damage justifies the cost of control — the level you are testing against. In these plans it might be 3 larvae per soybean plant, 1.5 Colorado potato beetle larvae per plant, or 250 soybean aphids per plant. The SPRT slope sits near this threshold, so the stop lines effectively ask 'is the field clearly above or clearly below the threshold?'

Many insect counts follow a negative-binomial distribution because individuals cluster rather than spread evenly; k describes how clumped they are, with a low k meaning highly aggregated (many zeros and a few high counts) and a high k meaning more even. k shapes the SPRT lines — aggregated pests like corn borer egg masses (low k) need the plan to account for the clustering. Each plan carries a published k you can calibrate locally.

The plan's error rates set the confidence: alpha is the false-positive rate (treating when you shouldn't) and beta the false-negative rate (missing a treatable infestation). A treat decision carries roughly 1 minus alpha confidence, a don't-treat decision roughly 1 minus beta. With both at 0.1, decisions carry about 90% confidence, which the tool reports alongside the decision.

Yes, when the infestation is unambiguous. If the first several units are all well above the per-unit threshold the cumulative count crosses the treat line quickly; if they are all near zero it crosses the don't-treat line once that line turns positive. This early stopping is the whole point — sequential sampling spends effort only where the decision is genuinely close to the threshold.

Pick your pest plan, then scout unit by unit (plants, sweeps, leaves) entering each count as you go. After each one the tool tells you treat, don't treat, or keep sampling, and plots your running total against the stop lines. When it says stop, you have your decision at the stated confidence; record the sample number to track how much scouting the plan saved versus a fixed count.

No — it is designed to give the same decision confidence (set by alpha and beta) as a fixed plan, just with fewer samples on average. The error rates are built into the stop lines, so a treat or don't-treat call carries the chosen confidence regardless of how few samples it took. The saving is in effort, not accuracy; borderline fields still get sampled thoroughly because the count stays between the lines.

Use it with your local economic threshold and extension plan, not instead of them. The plans here are representative mid-points of published SPRT plans and the aggregation parameter k should be calibrated to your region and crop. Treat the tool as the decision engine once you have a sound threshold and distribution for your pest, and confirm marginal calls with experience and local advice.