All of Them Anyway
Looking through a handful of digits you are holding in mind costs you a fixed amount of time per digit. This page measures the rate, and then checks whether you stop when you find the thing.
The five pages before this one in the lab all measured the price of a decision, and every one of them measured something that happened between a thing on the screen and a finger.
This one measures something that happens with the screen quiet and your hand still: how fast you look through a few digits you are holding in your head.
b is a rate, in milliseconds per item
Digits appear one at a time. Hold them. Then one more digit arrives inside a box: was it in the list? Yes or no, fast, without getting it wrong. Sternberg ran this in 1966 and it has been the way to time this ever since.
Every reaction time here contains reading, deciding, moving a finger, your keyboard and your display. None of that can be separated out, and all of it sits in the intercept.
But all of it costs the same on a one-digit list and a six-digit list. It adds a constant, and a constant cancels out of a slope. So the headline number is clean of your hardware even though the reaction time underneath it is not. This is also why answering on a phone is fine.
The sensible way to search a list is to stop when you find the thing. Under that scheme a YES answer goes through half the list on average and a NO answer has to go through all of it, so the yes slope should be half the no slope.
Sternberg got two parallel lines. Which says you do not stop: you go through all of them anyway, every time, and only then look at whether anything matched. Checking one more item is cheap. Asking yourself after each one whether to stop is not.
yes against no, and where in the list the match was
Three blocks of lists of one to four digits. Then that data is closed and two models are fitted to it: a straight line, and a logarithm, which is the law the Hick and Hyman page in this lab measured on a task that looks almost the same from the outside. More things to choose among costs the log. More things to search through should cost linear.
Over one to four the two are nearly the same curve, which is why the last block is all lists of six. Both predictions are printed on the screen before it opens. Then it opens.
Only the digit inside the box wants an answer. On a phone the two buttons do the same job. Five blocks, about seven minutes, the first one unscored, escape aborts at any time. A run below 75 percent correct is refused rather than scored, because a rate fitted to guesses is not a rate.
Everything runs in your browser. Nothing is recorded. Nothing leaves the page.