Mock Screens
Economic Consulting Analyst Screen
A single 35-minute, 22-item sitting built from what economic-consulting analyst work actually requires - difference-in-differences, instrumental variables, omitted variable bias, and antitrust market-definition math, with no negative marking.
What is an econ-consulting-style analyst screen?
Economic consulting firms - the Cornerstone Research, Analysis Group, NERA, Bates White tier of the industry - build litigation and regulatory work around expert testimony, which means every number an analyst produces has to survive cross-examination. Analyst hiring at these firms screens less for whether a candidate can run a regression (any statistical package does that) and more for whether they can say, under pressure, exactly what assumption makes a given estimate valid and what would break it.
This is why causal identification - not description, not correlation - sits at the center of the screen. A difference-in-differences estimate, an instrumental variable, a regression coefficient burdened by an omitted variable: each is only as good as an assumption that cannot be directly tested (parallel trends, the exclusion restriction, no omitted confounder), and the process is built to find candidates who reach for that assumption automatically rather than treating the arithmetic as the whole answer.
The antitrust side of the same screen - HHI concentration math, market definition, the hypothetical monopolist test - reflects that a large share of economic consulting revenue comes from merger review and competition litigation. These questions test whether a candidate understands that the arithmetic (a sum of squared market shares) is only meaningful once the market has been defined correctly, since the same merger can look concentrating or benign depending on how broadly the relevant market is drawn - and market definition is an economic argument, not a calculation.
Because expert reports are adversarial by nature, opposing economists are paid to find the weak assumption in any analysis, so these screens deliberately build wrong answers that look reasonable to someone who has not internalized the underlying identification logic - a level gap mistaken for a treatment effect, a broader market mistaken for a more accurate one. Filtering at the hiring stage for candidates who are not fooled by those look-alike answers saves the firm from the same mistake showing up on the stand years later.
How it works
The sitting is a single section, "Causal inference & competition" - 22 items in 35 minutes, mixing numeric-entry and four-option multiple choice across three areas: causal identification (difference-in-differences, instrumental variables, omitted variable bias, parallel trends), and antitrust math (HHI concentration changes from a merger, the hypothetical monopolist / SSNIP test, and how market definition changes the answer). The instructions warn upfront that several items have a defensible-looking wrong answer, and that the distractors are themselves the test.
Items are served one at a time and the clock runs continuously for the full 35 minutes with no per-item limit. allowBack is false, so once an item is submitted or skipped it cannot be revisited and the palette cannot jump backward; flagging an item does not let you return to it mid-run, but does surface it - with the correct answer and explanation - on the marked script at the end.
The section ends when the clock hits zero or all 22 items are used, whichever happens first; anything left unanswered is simply scored as skipped.
How scoring works
Each of the 22 items is worth 1 mark, and the section's penalty is 0 - a correct answer scores +1, a wrong answer scores 0 (not negative), and an unanswered item also scores 0. The section's net score is correct minus wrong times the penalty, which with penalty 0 reduces to the raw count correct.
Since a wrong answer and a blank score identically, there is no scoring incentive to skip an item you have any opinion on. The review screen shows the score out of 22, percentage of maximum, and per-skill accuracy across the skills the items are tagged with - regression, selection-bias, combinatorics, and game-theory - built only from items you answered.
As with the other sittings, no percentile is reported against other candidates - the review screen states plainly that this assessment does not yet have a comparison population.
Causal inference & competition
For every causal-identification item, answer the identification question before touching arithmetic: what is the assumption that makes this estimator valid, and does the scenario in the stem satisfy it? A DiD numeric item is just two subtractions, but the choice items around it test whether you know the assumption is parallel trends (not equal levels), and that a shock hitting only the treated group during the post period is the one thing that survives both differences and contaminates the estimate.
On the HHI items, compute the delta directly from the formula (Delta-HHI = 2 x s1 x s2 for a two-firm merger, since every other firm's squared share is unchanged) rather than reconstructing the whole index before and after - it is faster and it is the exact quantity being asked for. Then treat the market-definition items as a separate, non-computational question: a broader market dilutes every share, so Delta-HHI falls quadratically, which is why parties to a merger argue for the broadest defensible market and why market definition, not the arithmetic, decides these cases.
Every conceptual item in this section is built around one specific place where a superficially reasonable answer is wrong: omitted variable bias takes the sign of two products, not one intuition; a hypothetical monopolist test failing means the candidate market is too narrow, not that the merger should be blocked. Read each stem for which specific mechanism is being probed rather than pattern-matching to the general topic.
Pacing across the whole sitting
35 minutes across 22 items is about 95 seconds each on average, but the numeric items (DiD, HHI delta) take under 30 seconds once you know the formula, so budget the time you save there for the conceptual items, which reward reading the stem twice to catch which specific assumption or mechanism is being tested.
Because allowBack is false, flag anything you are unsure of and move on rather than re-reading a stem three times in place - a flagged item gets its explanation on the marked script afterward regardless, while time spent stalled on one item is time the clock is taking from the rest of the 22.
With no negative marking, submit your best read on every item before the clock reaches zero. An educated guess between two defensible-looking choices costs nothing next to a blank, so there is no scoring reason to leave anything unanswered.
A worked example
The item's table shows a treated group (A) and a control group (B), each with a before and an after figure.
Take each group's own before/after change first.
The DiD estimate is the difference between those two changes - the treated group's change net of whatever the control group would have done anyway.
Enter -70. The number is negative because the treated group fell further than the control - the estimated effect is a decline of 70 units beyond the common trend, not a fixed level gap.
Common mistakes
• Computing the treated group's raw change and stopping there, without subtracting the control group's change. That single number is not a DiD estimate - it does not net out whatever would have happened anyway.
• Treating equal pre-period levels as the identifying assumption for DiD. The assumption is parallel trends - equal levels are neither required nor sufficient.
• Assuming a broader market definition makes an HHI estimate more accurate. It mechanically shrinks every party's share and drives Delta-HHI down regardless of the true economics, which is exactly why market definition is argued, not computed.
• Getting the sign of omitted variable bias backward. The bias takes the sign of (effect of the omitted variable on the outcome) times (its correlation with the included regressor) - work out both signs before concluding the direction.
• Leaving a conceptual item blank because none of the four choices feels certain. With penalty 0, an educated elimination beats a guaranteed zero.
Why interviews test this
Case interviews and technical screens at economic consulting firms almost always include a DiD or IV scenario, because it is the fastest way to check whether a candidate reasons about identification rather than just computation - can you name the assumption, and can you say what evidence would support or undermine it? This format drills exactly that reflex: subtract correctly, then justify why the subtraction is valid in this scenario and not in a slightly different one.
The HHI and market-definition items map directly to how antitrust economists are actually questioned in depositions and expert reports: opposing counsel does not dispute that 2 x s1 x s2 is the formula, they dispute whether the market was drawn correctly in the first place. A candidate who can compute Delta-HHI instantly but cannot explain why the SSNIP test result changes the answer is missing the half of the job that survives cross-examination.
Play Economic Consulting Analyst Screen · All game guides · The arcade