How Many Hours a Day Should You Study for CAT? I Logged 335

There is no correct number of hours, and the number you have probably been given is arithmetically impossible for your week. Mine was: I set a weekly goal of 1,260 minutes - 21 hours - and my actual average across 135 active days is 149 minutes per active day. Dividing one by the other, hitting that goal needs 8.5 active days a week. There are seven. I ran that target for months without noticing, because I was watching the weekly total and never the ratio. Meanwhile the research is much less certain than the 10,000-hour story implies: in a preregistered replication of the study that started it, the best violinists had accumulated 8,224 hours by age 18 against the merely good group's 9,844 - a difference in the wrong direction, and not significant (p = 0.364).

So the useful question is not "how many hours". It is "how many hours can I put in at an intensity that produces a score", and the honest way to find that is to measure your own two numbers rather than adopt someone else's total.

The arithmetic that broke my own goal

Here is my log against the target I set for it. Everything in this table is first-party, from my own tracked CAT preparation.

Two ways out of that contradiction. Either the daily intensity rises - 1,260 minutes across seven days is 180 minutes every single day, with no rest day at all - or the goal comes down. I set 21 hours because "three hours a day" sounded like a serious number. It was a number about my self-image, not about my calendar, and the log is what made that visible. The full-time-work post covers the fragmentation that produced a 59-minute mean session in the first place.

What 335 hours actually bought

Across 20 full mocks between 10 May and 8 August 2026, my overall score went from 41 to 91 out of 198. That is the headline, and it is the least useful cut of the data. Two better ones:

The percentile column is where the hours show up. The section columns are where the honest lesson is: the same 335 hours produced 8.8 marks in QUANT and 2.2 in DILR. Hours are not a currency that converts at a fixed rate. They convert at a rate set by what you spend them on, and my DILR spend was the least well designed of the three. The full log, crashes included, is in the 20-mock post.

If you insist on a per-hour figure: 335 hours moved my mean percentile 20.9 points, which is about 16 hours per percentile point. Do not use that as a forecast. Percentile is bounded at 100, returns compress hard near the top, and the next 20 points are not available at any price. It is a description of one stretch, not a rate.

The 10,000-hour claim did not survive its own replication

The hours framing traces to Ericsson, Krampe and Tesch-Romer's 1993 study of violinists at a Berlin music academy, popularised as the 10,000-hour rule. Macnamara and Maitra reran it in Royal Society Open Science volume 6, article 190327, in 2019 - preregistered, double-blind, with 39 violinists in three skill groups against the original's 30.

What replicated and what did not:

• Did not replicate: the core claim that accumulated practice tracks skill among the elite. By age 18, the best violinists had logged 8,224 hours, the good violinists 9,844. That gap is in the wrong direction and is not significant (p = 0.364). In Ericsson's original, best had 7,410 against good's 5,301, significant at p < 0.05.
• Did replicate: good violinists had practised far more than the least accomplished group - 9,844 against 4,558 hours, p = 0.005. Practice clearly separates competent from poor.
• The ranges overlap enormously. Best violinists ranged from 3,978 to 14,664 hours; good violinists from 3,120 to 21,268. Somebody in the best group had done less than a fifth of what somebody in the good group had done.
• By age 20, both groups had passed 10,000 hours, which makes the round number useless as a threshold in either direction.

The authors state their own limitation clearly: 39 participants is still small, even at 33% larger than the original, and they cannot rule out that their violinists were at a different overall level than Ericsson's sample.

The controversy you should know about before quoting any percentage

The most-cited challenge to the hours story is Macnamara, Hambrick and Oswald's 2014 meta-analysis in Psychological Science, which is where the "deliberate practice explains only a small share of the variance" line comes from. I could not open that paper - it is paywalled, and I do not cite what I have not read.

What I could read in full is Ericsson and Harwell's open-access rebuttal in Frontiers in Psychology 10 (2019), which reports the meta-analysis's figures in order to dispute them. As reported there, deliberate practice accounted for 14% of performance variance overall: 26% in games, 21% in music, 18% in sports and 4% in education. Take those numbers as second-hand and contested rather than settled, because that is exactly what they are.

Ericsson and Harwell's objection is about definition, and it is a fair one. Ericsson's 1993 concept required individualised instruction from a qualified teacher, explicit goals, immediate feedback and repeated attempts at the thing you are bad at. The meta-analysis, they say, counted "engagement in structured activities created specifically to improve performance", which swept in group team practice, lecture attendance, general self-directed study, and in at least one case watching sport on television. Restricted to studies that met three stricter criteria, their reanalysis puts the share at 29% of variance, rising to 61% after correcting for measurement error.

You now have a range from 4% to 61% for "how much do hours explain", depending on whose definition of an hour you accept. That is the actual state of this field, and it is why nobody can hand you a number of hours with a straight face. What both sides agree on is the part that matters for CAT: the hours have to be aimed at something you cannot yet do, with feedback, or they are not the thing being measured. Ericsson and Harwell concede in the same paper that beginners practising alone often improve not at all, because they cannot see their own errors.

The 854,064-player study, and the two things it found besides volume

Stafford and Dewar, Psychological Science 25(2):511-518, analysed 854,064 players of an online game demanding rapid perception and decision-making, with a complete record of every play. It is the largest clean look at practice-and-performance I have read. Three findings, in descending order of comfort:

• Practice works. Average score rose with each consecutive play, holding for up to 100 plays.
• Spacing was worth about 50% extra practice. Among players of comparable starting ability, a 24-hour gap was worth around 3,000 points - "comparable to about 5 plays, in the 10-15 play range", in the authors' words. Spreading first and last plays further apart predicted higher scores (t(99) = 7.27, p < 0.0001).
• Volume was not what separated the top players. Grouping players by their eventual high score, "the difference in average score is present from the very first plays". The best players were ahead before they had practised, and improved faster afterwards. The authors note this "is in marked contrast to some popular and academic accounts of high performance which have denigrated the importance of talent with respect to practice."

That last one is unwelcome and worth sitting with. It also does not mean what a demotivated reader will take it to mean: the same effect held across the whole distribution, not just the top 20%, which means it describes a general gradient rather than a ceiling on you specifically. And it is a browser game measured over hours, not a three-year exam preparation measured over years - the transfer is a guess.

So what number should you actually pick

Stop picking a weekly total first. Pick the two numbers underneath it, because they are the ones you can control:

1. Minutes per active day. What does a real study day look like when it goes normally, not heroically? Mine is 149 minutes. Yours is whatever your last three weeks say.
2. Active days per week. Six is realistic for most people with a job or a college schedule. Seven is a plan to break the streak.

Multiply those two and you have a weekly goal that is arithmetically survivable. Mine, honestly computed, is 149 x 6 = 894 minutes, not 1,260. The 21-hour target was aspiration wearing a number's clothes, and a goal you miss every week stops carrying information about anything.

If you want a starting bracket rather than a derivation, here is the one I would defend for CAT specifically. These are my construction from my own log and the structure of the paper, not a research finding - no study has ever established an optimal weekly hour count for CAT.

Notice the last column. Every row has a cut order, because every plan meets a bad week, and a plan with no stated cut order gets cut in the wrong place - almost always at the full mock, which is the only item that produces the data telling you whether any of the rest is working.

Hours are the wrong thing to optimise once they are adequate

Three specific things beat adding hours, all with better evidence behind them than hour counts have:

• Spacing. Worth roughly 50% extra practice in the 854,064-player dataset. Same total volume, spread out. See the revision-schedule post.
• Mixing. A preregistered trial of 787 students found mixed practice beating topic-wise 61% to 38% on a delayed test - the interleaving post has the detail and the caveats.
• Analysis per mock. My 2.2-mark DILR improvement across 335 hours is not an argument for more DILR hours. It is an argument that the hours were shaped wrongly, which only a per-mock analysis habit can catch.

Where this is weak

• I have not sat CAT yet. Everything here is measured against mock percentiles, which are computed on the self-selected cohort that sat that specific paper, not on the roughly 2.5 lakh who sit the real thing.
• My log is n = 1, with no control. I cannot separate what the hours did from what three months of maturation, mock familiarity and paper-source changes did. Nobody can, from a personal log.
• The 2014 meta-analysis is paywalled and I did not read it. Its percentages appear here only as reported by its critics in an open-access rebuttal, which is a weaker basis than reading the original, and I would rather say so than launder it.
• Both sides of the deliberate-practice argument have a stake. Ericsson and Harwell are defending a definition they authored; Macnamara has published repeatedly against it. I have read one side in full and one side second-hand, which is not a balanced diet.
• Violinists and browser-game players are not CAT aspirants. Music is a domain with decades-long trajectories and continuous expert coaching. A browser game is measured in hours. CAT is neither.
• The bracket table is judgement, not evidence. There is no study establishing 12-18 hours for a working professional. It is the range my own log supports for one person with one job.
• The 16-hours-per-percentile-point figure is a description, not a rate, and it will not hold for the next 20 points because percentile compresses at the top.

Sources

• Macnamara, B. N., & Maitra, M. (2019). The role of deliberate practice in expert performance: revisiting Ericsson, Krampe & Tesch-Romer (1993). Royal Society Open Science, 6, 190327. Read in full via PMC. All violinist figures above are from that text.
• Ericsson, K. A., & Harwell, K. W. (2019). Deliberate practice and proposed limits on the effects of practice on the acquisition of expert performance. Frontiers in Psychology, 10, 2396. Read in full via PMC. The 14%, 26%, 21%, 18% and 4% figures are the 2014 meta-analysis's, as reported and disputed there.
• Stafford, T., & Dewar, M. (2014). Tracing the trajectory of skill learning with a very large sample of online game players. Psychological Science, 25(2), 511-518. Read in full via the White Rose postprint.
• Macnamara, B. N., Hambrick, D. Z., & Oswald, F. L. (2014). Deliberate practice and performance in music, games, sports, education, and professions: A meta-analysis. Psychological Science, 25(8), 1608-1618. Paywalled; not read. Listed because its numbers appear above, second-hand.
• Session log (340 sessions, 20,100 minutes, 135 active days) and mock log (20 attempts, 10 May to 8 August 2026) - first-party, from my own Karma Yogi account.

The one thing I would change about my first six months is measuring minutes per active day before setting a weekly target, instead of the other way round. Karma Yogi computes both, so the moment your goal becomes arithmetically impossible you can see it in a week rather than a quarter.