trialGroup

Experiments Surveys & forms

Caution

Legacy column. It still works in old experiments, but do not use it in new ones. Use trialOrder specs instead (for example 1: sample 10 difficulty keeps 10 trials per value of difficulty).

Modifier column for randomPick: its labels split a run of candidate rows into separate pick-pools, so each labelled group is sampled on its own. Use it when one contiguous run of rows that share the same type, randomPick, randomBlock and random values should be sampled as several independent pools instead of one.

The column has two effects. Its per-trial value decides where one pick-pool ends and the next begins. And the mere presence of the trialGroup header changes picking for the whole file: each pool presents the trials it keeps in a random order, even when nothing was dropped. In both cases the number picked is capped at the size of the pool, and rows with an empty randomPick keep their relative positions.

Syntax

Put a label in the cell of every row of a group, for example easy on one run of rows and hard on the run below it.

Any token is accepted, numeric or text, and the value is never interpreted: it only ever gets compared with the neighbouring rows’ values. A pool is the longest run of consecutive rows whose type, randomBlock, random, randomPick, trialGroup and timer label are all identical. A change in any of those, or a row with an empty randomPick, starts a new pool.

Options

Value	Effect
empty cell	Default. Empty still counts as a value: a change from empty to a label, or back, splits pools like any other change
any token (easy, 1, A)	Distinguishes this row’s pool from adjacent rows with a different token

There are no reserved values and nothing is validated: any text is accepted as typed.

Defaults & missing values

• If the column is missing, or a cell is left blank, that row carries no group label, which still separates it from neighbouring rows that carry one.
• Unlike random and randomPick, trialGroup values are kept on staircase trials.

Caution

The header alone is enough. As soon as a trialGroup column exists (matched case-insensitively), randomPick switches to the grouped mode for the whole file, even when every cell is empty.

Works with

• randomPick: the entire purpose of the column. With no randomPick value anywhere in the file, nothing is picked and trialGroup does nothing.
• type, randomBlock, random, timer labels: these also define where a pool ends, so a change in any of them splits pools even when trialGroup stays the same.
• random and randomBlock after picking: both run once the picking is done, shuffling within blocks first and then the blocks themselves.
• trialOrder specs (trialOrder: "N: sample ..."): a file cannot use both. Combining trialGroup labels with the specs is rejected when the trial file is saved; delete the trialGroup values and express the sampling with sample, select or groupSample in the trialOrder column instead.

Examples

Two labeled pools, 2 picked from each (both runs share randomPick=2, so without trialGroup they could merge if type/randomBlock/random also match):

type	stim1	stimFormat	randomPick	trialGroup
test	e1	.png	2	easy
test	e2	.png	2	easy
test	e3	.png	2	easy
test	h1	.png	2	hard
test	h2	.png	2	hard
test	h3	.png	2	hard

Fixed trials keep their position between pools (only pool members are ever dropped, so rows outside a pool keep their slots; here the break still lands between the A pick and the B pick, though it moves earlier in the sequence because each pool keeps only 1 of 2 trials):

type	stim1	stimFormat	randomPick	trialGroup
test	a1	.png	1	A
test	a2	.png	1	A
instructions	break
test	b1	.png	1	B
test	b2	.png	1	B

Tips & gotchas

Caution

The same label does not pool non-adjacent runs. A new pool starts wherever the grouping changes, and pools never merge back, so A ... B ... A gives three pools and the two A runs are sampled independently of each other. Grouping only ever applies to consecutive rows.

• Merely adding the column header, even with every cell empty, changes randomPick behavior for the whole project, because the switch looks at the header rather than at the values.
• The trials a pool keeps are shuffled among that pool’s slots, so per-pool order is randomized even without the random column, and even when you pick as many trials as the pool holds.
• trialGroup appears in the results file like any standard column.