SHL Verify Tips: Numerical, Verbal, Inductive and Deductive

SHL runs the most widely used aptitude tests in the world, which is exactly why most advice about them is useless. Telling someone to stay calm and read carefully does not help. What helps is knowing how much time each test actually gives you per question, and which specific mistake each section is built to catch.

What you are likely to be sent

Employers configure the length, so your invitation may differ by a few minutes. The ratios matter more than the absolute numbers: numerical gives you close to twice the time per item that verbal does, and candidates who pace both the same way run out of time on one and leave marks on the table in the other.

Numerical: the arithmetic is not the test

The maths in SHL numerical reasoning is GCSE level. Percentages, ratios, growth rates and simple proportions. What is hard is pulling the right three numbers out of a dense table in the thirty seconds before you can start.

• Read the question before the data. The table usually contains four times more information than any one question needs. Reading it first means reading most of it for nothing.
• Check the units line every time. Figures in thousands next to figures in millions is the single most common planted trap, and it produces an answer that is in the option list, because the wrong-by-a-factor-of-1000 value is usually one of the distractors.
• Percentage change and percentage points are different questions. A margin moving from 4% to 6% is a rise of two percentage points and a rise of fifty percent. Both will be offered to you.
• Estimate before you calculate. When the options are spread widely, a rough order-of-magnitude answer eliminates three of them in five seconds and you only need to distinguish the last two.
• Use the calculator properly. It is allowed, and on a test where the reasoning is easy, most wrong answers come from mental arithmetic slips or mistyped digits rather than from misunderstanding. Type the whole expression rather than transcribing intermediate results.

Verbal: Cannot Say is a rule, not a feeling

The verbal test gives you a passage and a statement, and asks whether the statement is True, False, or Cannot Say based on the passage. The three verdicts have precise definitions:

• True: the passage alone makes the statement follow.
• False: the passage alone contradicts it.
• Cannot Say: the passage does not settle it either way.

Cannot Say does not mean probably true, and it does not mean you are unsure. It means the text is silent. The failure mode is importing knowledge: a statement can be entirely true in the real world and still be Cannot Say here.

Three specific patterns account for most of the losses:

• Quantifier swaps. The passage says many customers, the statement says most customers. Many does not establish most, so it is Cannot Say.
• Causation smuggled into correlation. The passage reports that two things rose together, the statement says one caused the other. Cannot Say.
• Comparisons the passage never makes. The passage gives one region's figure, the statement claims it is the highest. Unless the passage says so, Cannot Say.

Inductive: check one dimension at a time

Inductive items ask you to find the rule governing a sequence of shape frames. The instinct is to stare at the whole frame and wait for the pattern to appear, which works for easy items and fails completely once two rules run at once.

The reliable method is to scan a single dimension across every frame before you look at the next dimension. The dimensions in use are almost always drawn from a short list: shape, count, size, shading, rotation, position, and the addition or removal of an element. Walk them in a fixed order. Once you have found one rule that holds across all frames, look for a second one before answering, because compound rules are what the harder items are made of.

Deductive: plausibility is a trap

Deductive items give you premises and ask whether a conclusion must follow. The premises are true for the purposes of the question even when they are absurd, and a conclusion that is obviously true in the real world does not follow unless the premises deliver it.

Two errors are worth drilling until they feel wrong:

• Reversing a conditional. All accountants are numerate does not give you all numerate people are accountants.
• Negating a universal too hard. The opposite of all managers attended is not no managers attended. It is that at least one did not.

Also remember that some means at least one and is compatible with all. A conclusion saying some X are Y follows perfectly well from a premise saying all X are Y.

Verify G+ is supposed to get hard

Verify G+ is adaptive. Answer correctly and it hands you a harder item, so the test walks you up the difficulty curve until you start missing. That means two things. Your percentage correct is not informative, because the test is engineered to push everyone towards missing about half of what they see near their ceiling. And the moment it stops feeling comfortable is the moment it has found your level, not the moment you are failing.

Candidates who read escalating difficulty as a sign they are doing badly start rushing, which drops them back down the curve. The correct response is to keep the same pace you started with.

The OPQ32 and answering as yourself

The OPQ32 is a forced-choice personality questionnaire: from each block of statements you pick the one most like you and the one least like you. There is no time pressure and no right answer, but the format is designed so you cannot be maximal on every desirable trait, because every block makes you give something up.

Attempting to answer as an idealised candidate produces a profile that is internally inconsistent, and inconsistency across a questionnaire this long is visible. Answer as yourself at work rather than as yourself generally, and answer quickly. First-instinct answers produce a more coherent profile than deliberated ones.

The verification retest

The Verify range was built around the idea that an unsupervised test can be confirmed later. Some employers follow the online test with a shorter supervised version, and where they do, a large gap between the two is the result that matters. That is the concrete reason not to take help on the unsupervised sitting: help does not raise your score, it just moves the failure to a later stage where it is more visible.

What practice actually changes

Practice does not make you smarter, and anyone selling that is selling nothing. What it removes is decoding time. On a first sitting, a large share of your 75 seconds goes on working out what the question wants and where the data lives. On a fifth sitting that part is close to free, and the whole budget goes on the reasoning. That is where the score comes from. You can run timed simulations of every test in the Verify range here, with the same clocks the real tests use.