Reasoning6–10 minBurden: HighEMA: Baseline / infrequent

Number Series

A row of 5–7 whole numbers follows a rule; the participant types the number that comes next on an on-screen number pad. Items are generated from a seed at four rule levels. Measures numerical inductive reasoning.

Inductive reasoningNumerical reasoningFluid reasoning
Category
Reasoning
Typical duration
6–10 min
Participant burden
High
EMA suitability
Baseline / infrequent

Try this task

Configure parameters and run an interactive preview exactly as participants will experience it. No data is recorded.

Configure preview

Adjust parameters below, then start the preview on the right.

Include practice trials

Shown with feedback before the main task

Task parameters

Seed: empty gives every run its own items; the seed drawn is stored with the result (seed), so the items can be rebuilt. A whole number (0–4294967295) gives every participant the same items.

Rule levels

20 items: 5 at level 1, 5 at level 2, 5 at level 3, 5 at level 4, in ascending order of level; within a level the rule families are balanced and in random order. Practice: 2 items at level 1, with feedback. 30 s per item; about 8 min.

Time limit for the whole block: off by default, so each row has its own limit. The published number series tests limit the whole set instead: 10 min for 20 items in the I-S-T 2000 R and the WIT-2 (Schneider & Sparfeldt, 2021) and in the CogAT (Bermingham et al., 2013); to give the task that way, set 10 min here and the time limit per item to 0. With a block limit, the rows come in order without going back, a countdown is shown, the row on screen when the time is up is kept but not scored, and rows never shown are reported as not reached. Practice is not timed.

Defaults: 20 items with 30 s each. The number series subtests of the I-S-T 2000 R and the WIT-2 have 20 items in 10 min (Schneider & Sparfeldt, 2021), as does the CogAT number series (Bermingham et al., 2013); those tests limit the whole set, so a limit per item is an adaptation to one item per screen. The rule families follow the number series item models of Loe et al. (2018) and the analyses of series completion by Simon & Kotovsky (1963) and Holzman, Pellegrino & Glaser (1983). The number of practice items, the largest number (999) and the blank between items (500 ms) have no published value (our choice).

Every item is generated from the seed and checked before it is shown: no rule of a lower level fits its numbers, and every rule of its own level that fits gives the same next number. Rows show 5–7 whole numbers from 1 to the largest number, so answers are never negative. Note: Loe et al. (2018) found that a change that itself changes (level 2 here) made items harder than interleaved series (level 3), so the levels order rule types, not measured difficulty.

Outputs: correct answers; answered items (✓ before the time ran out; a row whose time ran out counts as not answered, even with digits typed); accuracy per rule family; median time to a correct answer; with a block limit, the rows not reached. Per item: level, rule family, the numbers shown, the answer, the typed response, the time from the drawn row to ✓ and to the first key, and the number of deletes.

6–10 minBurden: HighEMA: Baseline / infrequent

This is a researcher preview. No participant data is recorded.

Simulated participant view

9:41

Number Series

A row of 5–7 whole numbers follows a rule; the participant types the number that comes next on an on-screen number pad. Items are generated from a seed at four rule levels. Measures numerical inductive reasoning.

No data is recorded

Participant experience on smartphone

One row at a time: the numbers in boxes, a dashed box for the missing next number, and a number pad below with delete and ✓. A thin bar shows the time left for the row. Practice rows say whether the answer was right and give the correct number.

When to use

Useful in baseline or occasional sessions that need a numerical reasoning measure with many parallel forms, for example where number series as used in aptitude and selection tests are practised or tracked over time. Each run can get new items (no seed), or every participant the same items (a fixed seed).

When not to use

Not a normed test and not a substitute for the number series subtests of standardised batteries: the items are generated, and their difficulty has not been calibrated. Not a quick EMA probe. Less suitable where arithmetic skill or math anxiety would confound the target construct.

How to use in a study

Keep the defaults (20 items, 30 s per item, all four levels in ascending order) for a form close in length and pace to the published 20-item, 10-minute number series subtests. For repeated sessions leave the seed empty, so each run gets new items; set a seed when all participants should see the same items. Report accuracy per rule family or level and use the item-level data, not speed alone.

Researcher-configurable parameters

  • Number of items: 20 (the number series subtests of the I-S-T 2000 R, the WIT-2 and the CogAT have 20 items)
  • Time limit per item: 30 s (those tests allow 10 min for 20 items; a limit per item is an adaptation to one item per screen), or no limit
  • Time limit for the whole block: default off; 10 min for 20 items as in the I-S-T 2000 R and WIT-2 (set the limit per item to 0 to give the test as published). Rows not reached are counted
  • Rule levels 1–4 (default all): constant difference; constant factor or differences that grow by a constant; two alternating operations or two interleaved series; a rule in the differences or each number the sum of the two before
  • Practice items: 2 at the lowest selected level, with feedback (no published value)
  • Largest number shown or asked: 999 (no published value; Loe et al., 2018, advise against overly large numbers)
  • Seed: empty (new items per run, the seed is stored with the result) or a fixed number (the same items for everyone)

Outputs collected

  • Correct answers and answered items (✓ before the time ran out)
  • Accuracy per rule family
  • Median time to a correct answer
  • Every item: level, rule family and parameters, the numbers shown, the answer, the typed response, the time from the drawn row to ✓ and to the first key, the number of deletes, and whether the time ran out
  • The seed, so the items of a run can be rebuilt

Interpretation notes

Each item is checked when it is generated: no rule of a lower level fits its numbers, and every rule of its own level that fits gives the same next number, so each item has one answer under the rule families used here. The levels order rule types; they are not calibrated difficulty steps. Loe et al. (2018) found that a change that itself changes had the largest effect on item difficulty and parallel sequences the smallest, so level 2 items with growing differences may be harder than some level 3 items with interleaved series. A row whose time ran out counts as wrong and as not answered, even if digits were typed. The published tests set one time limit for the whole set and allow going back; here each row has its own limit and rows cannot be revisited. As in the I-S-T 2000 R, the answer is typed, not chosen from options. All numbers are whole numbers from 1 to the largest number. Responses are timed from the drawn row to the touch-down, but this timing has not yet been validated on real devices.

Scientific evidence

  • Simon and Kotovsky (1963) developed a theory and descriptive language for letter series problems; Holzman, Pellegrino and Glaser (1983) extended the scheme to number series, with four components: relations detection, discovery of periodicity, pattern description and extrapolation (as summarised by Loe et al., 2018).
  • In number series, process and content-knowledge variables accounted for more than 70 percent of the variance in item difficulty, in children and in college adults (Holzman, Pellegrino & Glaser, 1983).
  • For number series items generated automatically from rule types, the item difficulties predicted from the rule types (linear logistic test model) correlated r = 0.83 with the Rasch difficulties (Loe et al., 2018).

Timing and session conditions

Each run of this task records how its times were measured and the conditions of the session, such as screen brightness, battery saving and interruptions. Runs with possible problems are flagged on the Quality tab and in the export.

How SMAAT times cognitive tasks

Read more

Related tasks

Add Number Series to your ESM study on SMAAT — free to get started.