Search

• A basis is a subspace
• When the constraints run out
• Where the guarantee is exactly zero
• Which shapes are worth protecting
• What the ellipse spends its assignments on
• A box with a count

• A block size that changes
• A schedule that reads the mean
• Two degrees of freedom, one total
• What a schedule actually buys
• How much of the forbidden spread a rule may read
• A rule that sets its own κ

• A block weighted inside itself
• A taper and a critical value
• A block at every starting row
• A length chosen for the interval
• A share read off the sample
• A sequence that knows it was centred

• A break that was looked for
• A second break on a flat profile
• Choosing whether to break
• What a search costs in parameters
• The break a shared scale hid

• A charge that depends on the rule
• The charge that is not a sum
• Two searches, one sample
• A break that is there
• A set calibrated at the estimate

• A copula that halves a marginal
• A symmetry that was not enough
• The zero that survives both
• Two failures that cancel
• A median with ties
• The zero that needed two

• A count that has to be estimated
• A dictionary that is a product
• What a reference distribution costs to sample
• Where the enumeration stops
• A basis drawn rather than chosen
• A floor for the draw

• A covariance with no parameter in it
• Errors generated from a fitted model
• The triangle that was not the multiplier's
• The window that has to be chosen, and the term that was dropped
• A covariance repaired

• A covariate with no levels
• Balancing more than one number
• The rule that reads the number
• What the balanced trial is worth
• A number read with error
• A second reading or a second covariate

• A criterion is a prediction of the hold-out
• How long a block a multiplier shares
• Two defects and one resampling
• Where the two searches cross
• A penalty that counts independent rows
• A charge the sample can afford

• A cut is not a polynomial, and it does not have to be
• The arcsine that closes it, and the error that was overstated
• The zero that survives a cut
• A rate needs a rate beside it
• A companion cut into many
• A companion with two coordinates
• Weights from an approximate law

• A defect that is about size
• A test rather than a survey
• The diagnostic at two hundred
• The statistic that changes sign
• A unit that has not moved

• A dependence fitted with the line
• A window for every candidate
• Iterating is not maximising
• The fit that takes the memory out
• A fixed point with no likelihood under it

• A dependence with a shape
• The order the tail is drawn at
• The window a whitening wants
• Where the generality runs out
• The residue was the weights

• A design is a number
• Four letters and two camps
• The design that has to be integers
• The theorem that says when to stop
• How many places a design goes
• Augmenting a design that has already run
• The criterion with no derivative
• A certificate for the worst direction

• A dictionary that is neither
• The set a dictionary leaves
• The walk that cannot cross
• What the extra function buys
• Two functions a covariate

• A distribution drawn from the null
• Eight forecasters and one benchmark
• The models that were never in the running
• What the other forecast adds
• The threshold the recentring reads
• A threshold against its sample

• A family before a fit
• Nothing in the fit picks the width
• The plug-in and the maximum
• What fitting them together buys
• An interval that knows the band was maximised

• A lag the sample has less of
• A line that beats two curves
• The width a band is measured in
• What a better charge buys
• A charge that reads the draw
• What the correction assumes
• The surface under every law

• A length for each instrument
• The instrument and the reading
• The reversal that was the instrument's
• What the interval covers
• A shrink towards the answer

• A list is not a rule
• The comparison that was not made
• The volume a whitening moves
• Two values the criterion cannot tell apart
• A quarrel's cost is its width

• A margin that turns over
• An answer that changes
• Where the skew puts a threshold
• The other dial
• The zero that was a crossing
• A cut on each side

• A model and a count
• A quantity that loses to a heuristic
• Counting it exactly does not help
• What the rule blocks
• A set of pairs, not a vector
• How far, not whether

• A null with a model in it
• A table of nested models
• The weight that is a vector
• When every null is true
• One test for eight nulls

• A penalty is a trace
• One number for a table of candidates
• The repair that was exact and made it worse
• The residuals are not the errors
• What a multiplier cannot keep

• A prior on the spread
• The interval that integrates
• What the plug-in forgets
• When the spread estimates to zero
• A control borrowed from the last trial
• A history that agrees with itself
• A history that has been moving
• A history that stepped

• A probe chosen from the design
• The part the rule already took
• What a chosen probe finds
• How far to project
• A ceiling that is one unit

• A probe nobody chose
• Before the trial and after
• Half a reference distribution
• The statistic the p-value is about
• After the mirror is gone

• A proposal that moves more than two units
• A ratio that changes between blocks
• Stationary is not convergent
• The bias that lands in the slope
• Where the gain is, and where the decision is

• A rate times a size
• How often it matters
• The quarrel that changes the winner
• What a longer sample settles
• The table decides when the sample settles it

• A search that is already the other
• Three quarters of the way to one search
• Two searches that share nothing
• The overlap at the critical value
• The power a calibration buys

• A split survives what a mean does not
• A zero that rests on a symmetry
• Balancing a skewed covariate
• The cut that is not a quantile
• A share at a hundred units
• The covariate on its ranks

• A split that depends on the order
• Two effects in one number
• What a zero is made of
• The split at the critical value
• A tail the law gives the pair

• A table and a list
• The eighth that was not a constant
• Two factors pointing opposite ways
• A step that is not a ratio
• The same dial under a second law

• A threshold in the tail
• Balanced on the wrong function
• The analysis and the shape
• Three functions of one number
• How many functions a trial can hold
• The list it was chosen on

• A width promised for a difference
• The degrees of freedom in the sums
• What a two-arm rule may not pool
• Which weights are the inverse variances
• The sums made independent
• The sums under a moving level

• A width that moves and an error that does not
• The charge nobody derived
• What a window leaves free
• The charge at four times the sample
• A cap that decides nothing

• A width the trial has to stop for
• Stopping on the arms
• The trials that stopped early
• A width rule on skewed outcomes
• A weighting that does not read its own spreads

• A zero that is arithmetic
• The symmetry the marginals could not show
• Which tail the cut sits in
• A cut walked into the tail
• A leak read off a two-by-two table

• A zero that was an assumption
• Draws that repeat each other
• The fourth moment that was missing
• Walking the admissible set
• The share a sample removes

• Allocating on a guess
• Not half and half
• One control, many arms
• The cost of a unit
• The arm whose variance is its answer
• Two contrasts, one split
• A contrast chosen after the data

• An efficiency that is a ratio
• Protecting one parameter over a range
• Stopping when it is precise enough
• The interval after a stop it chose
• The fifth a third stage never reaches
• A floor that reads the first stage

• An interval that carries its scale
• The ordering reverses again
• What studentising costs
• The count or the length
• The interval with no resampling in it

• An ordering that depends on the rule
• The length nobody has
• What choosing the length costs
• A length read off the whole sequence
• A ranking that belongs to the interval

• Balancing towards unequal targets
• Guessing one arm in three
• The analysis after three arms
• Three arms and three scores
• A guesser who sees the totals
• The guesser who learns

• Balancing what is known in advance
• The analysis has to know the rule
• The reference the covariates supply
• The rule that can be guessed
• The interval the test inverts to
• An effect that is not one number

• Bias is not the whole of it
• Measuring a variance rather than a quantile
• The error no window repairs
• The gap a sample shows
• A model against a window

• Blinded, and still exact
• The condition that cannot be dropped
• Weights that need only a ratio
• A ratio that changes with the period
• A stopping rule that reads the periods

• Borrowing towards a line
• Pooling a proportion
• The slope that borrows
• Two levels at once
• Two groupings that cross
• A level with two units
• What a two-unit study should report
• A prior on the spread between two sites

• Choosing n after looking
• Dropping the losers
• Randomising towards the winner
• The estimate after the choice
• Shrinking the arm that was chosen
• A prior that lets one arm stand out
• A second stage sized by the first
• Patients spent before the choice

• Choosing the order
• The interval after the choice
• The interval that forgets it estimated
• What the model says next
• Shocks that are not normal
• A tail sixty residuals have barely seen

• Correcting the persistence
• The repair that moves the wrong number
• When one model contains the other
• Which forecast is better
• Correcting the forecast instead
• What the interval is short by
• The correction that leaves the region
• An interval built from simulated futures

• Counting what is still wandering
• Three series and a count
• Which series does the moving
• Which series goes on the left
• Which mistake about the rank costs
• A space is not a relation
• The rank is a decision
• A restriction the data can check

• Eight groups, one population
• The prior the data estimates
• The weight that decides
• When borrowing goes wrong
• The fewest groups that can borrow
• Where the borrowing goes
• One population, or two
• The shape the estimates keep

• Four datasets, one summary
• R² is not a measure of fit
• Twenty residual plots
• Two points that hide each other
• The line that one point drew
• A robust loss and a far x
• The start an efficient robust line inherits
• The other half a trimmed fit rejected

• How many subjects
• One factor at a time
• Randomisation is not balance
• The variance removed before the data
• The word a fraction costs
• The design that refuses the corners
• The run that did not happen
• A run lost for a reason

• More data is not monotonically better
• The shortest interval is the one that misses
• The correction for not knowing the spread
• Twenty intervals and one expected miss
• What the 95% refers to
• Where the bootstrap lies
• An interval that covers and says nothing
• A count that bets against its interval

• Regression to the mean
• Simpson's reversal is a region, not a table
• What a positive test is worth
• Two analyses of one baseline
• The measurement that got them enrolled
• A lead that a heavy tail keeps
• The slope of a density nobody can see
• A correction built on a retest

• Residuals that keep their own variance
• The corner the test is calibrated at
• The displacement is a parameter count
• When the benchmark is a candidate
• The displacement on a heavy tail
• An excess and its window

• Spending the error rate
• When the looking happens
• The effect a stopped trial reports
• The outcomes a trial could have stopped with
• A boundary for giving up
• A look the trend asked for
• A look added by a rule
• The variance that contains the effect

• The base rate was always Bayes
• What a credible interval covers
• What a prior is worth
• Where the two schools agree
• The shortest interval, and the one that does not move
• When the prior is confident and wrong
• An interval for something else
• A prior wrong about its own spread

• The check before the standard error
• The observations that repeat each other
• Two walks and a finding
• What differencing costs
• The repair that keeps the question
• How slow a return a sample can see
• The cliff that is a slope
• The interval the joint fit reports

• The cost of differencing a pair
• The model that corrects its error
• The regression that is not spurious
• The test with no table
• A relation supplied in advance
• A slow return across many pairs
• A shock every pair shares
• A factor estimated from the panel

• The data that stops early
• The curve that survives censoring
• The interval at the end of the curve
• A dropout the data cannot see
• One minus Kaplan–Meier is not a risk
• The hazard ratio the follow-up chose
• A horizon chosen after looking
• One test for a curve that might cross

• The design for the worst case
• The design that stops guessing
• What a design chosen from the data costs
• Where the minimum is attained
• A bootstrap that repeats the design

• The design that cannot see a curve
• The optimum is a ratio, and its interval is sometimes the whole line
• Three levels, and the ring where the design says the same thing
• Walking up the gradient
• The sign the curvature has
• When the best setting is outside the region
• The run that confirms it
• The height at the chosen setting

• The design that hedges
• The design that needs the answer
• The family behind the letters
• The two terms anybody wanted
• A subset the model cannot use
• A design chosen for a test

• The experiments that could have happened
• The plus one and the round number
• The test that needs the rule
• What the exactness buys
• The null the exactness is for
• A statistic that is exact twice
• The repair, with a covariate in it
• The covariate chosen afterwards

• The guess with two numbers in it
• The rule that cannot see the mean
• The worst case in two directions
• What the blindfold costs
• A variance that follows the response

• The price of control
• Two different promises
• What the correction corrects
• False discoveries that arrive together
• Estimating how many nulls are true
• An order that spends the error rate
• Intervals for the findings
• An interval that knows its finding was found

• The seed is part of the figure
• Two routes to every number
• The same draws for both methods
• A simulation that stops when it looks settled
• A coverage table with its own error
• The draws aimed at the tail
• The check worth more than the check
• A proposal fitted to its own draws

• The shape, and where its mass is
• The tail converges last
• What normal actually looks like
• A ratio whose interval has to be the whole line
• Sums of almost anything
• Where the derivative is zero
• A flat point with more than one direction
• Testing for a flat point first

• The winner's curse
• Twenty analyses of nothing
• What a p-value does not say
• The p-value a replication gets
• A p-value that is not flat is not a p-value
• Two ways to combine p-values
• The smallest of three combinations
• Studies that share a control

• One arithmetic, three decisions
• The two worlds that look the same
• Adjusting for everything
• The sample is a condition
• The variable the treatment caused
• A collider before the treatment
• Adjusting for a shadow
• A set of shadows

• Three shapes, one limit
• The maximum converges slowly
• The threshold is a dial
• A level with no data in it
• Two intervals for one return level
• The clustering the tail has
• The run length a declustering chooses
• Where the extremal index matters to a return level

• The assumption nothing tests
• Weak, and back where it started
• Whose effect it is
• Two instruments that disagree
• What the first stage does not know
• Leaving each row out of its own first stage
• A standard error that knows about the instruments
• Cells of unequal size and an error that follows them

• What a wrong model estimates
• The bread and the filling
• Robust is not free
• Three corrections and a leverage
• Right for the wrong reason
• The count that is not the rows
• The reference the sandwich is read against
• An effective number of clusters

• Coverage from exchangeability alone
• What the split costs
• Marginal is not conditional
• The score is the modelling
• When the order matters
• The weight that has to be estimated
• A detector built for the ordering
• A quantile over the recent scores

• Dropping the incomplete rows
• One imputation is not an observation
• The variance between imputations
• An imputation model the analysis does not contain
• The mechanism the data cannot see
• Three mechanisms and one dataset
• The assumption that identifies the mechanism
• A variable that moves recording

• An identity in three terms
• A curve that is a binning
• Calibrated and useless
• The miscalibration a perfect forecaster shows
• A score that rewards lying
• The liar with two answers
• A forecaster that rounds
• Five values chosen by the forecasts

• How many observations a weight leaves
• A score that balances
• The region with no comparison
• Either model, but not neither
• The estimated weight is the better one
• A weight fitted to balance
• The moments a balance is told
• A balance allowed a tolerance

• Two standard deviations of what
• A tenth as wide, and both of them right
• Ninety-three observations, and nothing assumed
• All of the next ten
• A band allowed a few misses
• A range allowed a few misses
• A range that grows with the lot
• A range that forgets

• The reversal a coin cannot prevent
• The change that is not confounding
• Conditioning on what the treatment caused
• Every patient at one half
• Twenty variables nobody stratified on
• Five places to cut one variable
• A cut that slides
• Every variable, every cut

• How many analyses there really were
• What naming it in advance costs
• The correction that makes the estimate worse
• An interval for the winner
• Naming a handful in advance
• An interval for the analyses admitted to
• A winner chosen by its diagnostic
• The cleanest of the significant

• R² is a property of the design
• The t statistic wearing different clothes
• The summary that was meant to work
• Two numbers for the fit's geometry
• The points a slope rests on
• A point ordinary on every axis
• A term built from the others
• A value read off a curve

• Where the two tails disagree
• A degrees of freedom that is not a count
• The side a bound is read from
• The skewness of a difference
• A limit from the family it came from
• The level a limit should be set at
• A failure charged by its size
• A level set before the source is known

• The second test that is not a second opinion
• The test is a point somebody chose
• The prevalence the test has to estimate
• An upper limit is the finding
• A threshold that jumps
• The threshold a validation study can see
• A test whose tail is not normal
• A threshold reported with its interval

• The plot is about the wrong quantity
• Residuals are not the errors
• The band the eye was standing in for
• Three runs at the end of the line
• A number for the shape
• Residuals that look too normal
• Residuals made independent
• Residuals of a model the data chose

• A bound written for a coin
• A correction that goes below zero
• An approximation built at the threshold
• A tail the sample never saw
• The same correction, inverted
• The correction a t test would use
• A tilt for the ratio

• A hole no sample size fills
• What a guaranteed minimum costs
• The coin that makes it exact
• The coin that is already there
• The coin a clock supplies
• Two windows and a ratio

• Two intervals that overlap
• Significant in one, not in the other
• Five times in six
• The verdicts watch the threshold, not the gap
• A subgroup inside its own trial
• Sixteen subgroups and one effect
• One characteristic that really matters
• The pattern a cause leaves on its neighbours

• A group from the population's own tail
• A league table of a hundred
• Estimates that are too alike
• The small groups one centre protects
• The relation the table has to estimate
• The interval a small group is given
• A ranking with its uncertainty
• A centre chosen by its own table

• The spread a pilot supplies
• The chance a trial succeeds
• An outcome cut in two
• A baseline cut in two
• A visit before or a visit after
• The cut that fitted best
• The correlation a pilot can see
• A schedule the trial chooses for itself