Normal Distribution
nothing is called 95% until it has been counted
The collection

Every essay — page 10

Essays 217 to 240 of 613, in the same order.

Coverage without a distribution

Every interval counted here so far has been found short of what it promised. This one is not, and the reason is that it is not a statement about the data at all: the rank of a future observation among a set of exchangeable calibration scores is uniform, so an interval built at the right order statistic covers a stated share of the time at any sample size, under any distribution, around any model however wrong. What it does not say is where that coverage sits — and everything the procedure does not know turns up in the answer to that rather than in the total.

The split decides the width. The width of the interval against the share of 200 observations spent on fitting rather than on calibrating, over 3000 draws. Spending more on the fit shrinks the residuals; spending more on calibration builds the interval at a less extreme order statistic. The two meet at 0.5, where the width is 4.0416 against 4.1603 at 0.1 and 4.3820 at 0.9. Full conformal, which spends the same 200 points on both jobs, is 3.9865 — so the whole cost of splitting is 1.38%.

What the split costs

Splitting a sample between fitting and calibrating looks like a trade against the guarantee, and it is not: coverage moves 0.63 points across nine splits and every reading sits on its own promise. The whole cost is 1.38% of width — and at sixty observations the width falls, rises and falls again.

6 figures · Exchangeability, part 2 One promise, kept on average and inside neither group. What each calibration scheme covers inside each of two equally common groups whose noise scales are 1 and 3, over 6000 draws with 200 calibration points. One interval for everybody covers 100.00% of the quiet group and 90.66% of the noisy one, averaging to 95.28% — and the closed form for that population says 99.9999% and 90.0001% at a half-width of 4.9346, from two normal cdfs and no simulation. Dividing by an estimated per-group scale gives 95.29% and 95.48%; calibrating separately inside each group gives 95.49% and 96.01% against a closed-form expectation of 95.4645%. Only the last of those is a guarantee rather than a repair, because the rank argument runs inside each group.

Marginal is not conditional

One exactly valid interval covers 100.00% of a quiet group and 90.66% of a noisy one, and the floor is arithmetic rather than a measurement — a group of share π is guaranteed only 1 − α/π, which is zero when the group is as rare as the miss rate.

6 figures · Exchangeability, part 3 Six scores, one coverage. What each nonconformity score's interval covers, over 2000 draws with 200 calibration points, against the 95.0249% the rank argument promises. The column runs from 94.30% to 94.90%, a spread of 0.60% against a standard error of a difference of 0.69% — one number, six times. That includes a score aimed five units off the fit and a score that never reads the response at all, because the rank argument does not read the score either: it needs the scores exchangeable and nothing else. Every decision a modeller makes has to show up somewhere else, and the next two readings are where.

The score is the modelling

Six nonconformity scores on the same draws cover within 0.60 points of each other, against a standard error of a difference of 0.69 — one number six times. Their widths run over a factor of 2.361 and their adaptivity over a factor of 8.377.

6 figures · Exchangeability, part 4 What a scale that grows across the sample costs. What the interval covers when the noise scale grows across the sample, against how far the departure has gone, over 1500 draws at each setting. The coverage runs from 94.47% at no departure to 83.93% at the end of the sweep, a loss of 11.07%. The rank argument needs the 200 calibration scores and the test score to be exchangeable, and this is one of the three ways that fails. A test built for it reaches 80% power at 4.054, where the coverage is 85.13% — so 9.87% of the loss is inside the region such a test would have missed.

When the order matters

Three ways of breaking exchangeability cost 4.93, 11.07 and 1.07 points of coverage, and the ordering by cost is the reverse of the ordering by how soon a test would have caught them. The departure practitioners check for is the cheapest one.

6 figures · Exchangeability, part 5 Too small breaks it and too large does not. Coverage of the weighted interval against the factor the true likelihood ratio is multiplied by, at a test population 80.0% drawn from the noisier group and 200 calibration points. The exact weight is the factor of 1 and covers 95.70%. Overstating it costs nothing: 96.13% at sixteen times too large. Understating it costs, and costs steeply below about a half — 94.93% at half, 88.37% at an eighth and 67.90% at a thirtieth. The question this answers was whether a wrong weight degrades smoothly or falls off a cliff, and the answer is that it does neither symmetrically: the curve is smooth and one-sided.

The weight that has to be estimated

A likelihood ratio sixteen times too large costs 5.5% of interval width and no coverage at all; one a thirtieth of the right size covers 67.90%. The estimate from a batch of five unlabelled covariates covers 95.10% against an exact repair's 95.30%, and the binomial says why.

4 figures · Exchangeability, part 6 Three detectors for one departure, all at 5%. How often each of three checks on the calibration scores fires, against the size of the drift, with every critical value simulated under no drift so that all three sit at 5.0% exactly. The incumbent — a rank comparison of the first half of the scores against the second — reaches four-in-five power at a growth factor of 4.31. Reading each score's rank against its position reaches it at 2.65, and the largest running departure of the scores from their mean at 2.12. The ordering of the three is the ordering by how much of the sample's arrangement each one uses.

A detector built for the ordering

The best of three checks for a drifting scale fires at half the growth factor the standard one needs — 2.12 against 4.31 — and still leaves 6.50 points of coverage gone before it does, against 0.51 for serial correlation. The reversal was not a property of the test.

5 figures · Exchangeability, part 7 Coverage under a drifting scale, with the quantile taken over fewer and more recent scores. The noise scale grows linearly across the sample to 1 + v times its starting value; 4,000 draws at each v. Worst coverage over the sweep: all 199 scores 85.63%, the last 99 90.88%, the last 39 93.42%, the last 19 93.53%. At v = 2.12, where the best of three detectors first fires 80% of the time, they cover 88.11%, 91.86%, 94.02%, 94.45%. Every window covers 95% when nothing drifts.

A quantile over the recent scores

A noise scale that grows across the sample takes a split conformal interval from 95% to 85.63%, and the best detector cannot see most of that loss. Take the quantile over the last thirty-nine calibration scores instead of all 199 and the worst coverage across the whole sweep is 93.42%, for 3.7% more width when nothing has drifted. The same window covers 91.30% where all 199 cover 93.53% under strongly correlated errors — and switching to it only when the detector fires is worse than either, because the detector fires on correlation too.

4 figures · Exchangeability, part 8

The observation that has not happened

Two fields here stop at estimation. A forecast is the other question — not what the parameter is but what the next observation will be — and the band round it is computed by substituting estimates into a formula derived for the truth. Counted, that 95% band is not 95%, the shortfall grows with the horizon, and the model has to beat two benchmarks that estimate nothing at all.

One forecast, and the band the arithmetic puts round it. An AR(1) with φ = 0.75, 60 observations, fitted by least squares and forecast 14 steps ahead. The point forecast decays towards the fitted mean at φ̂^h; the band is ±1.96 standard errors from σ̂²Σψ̂², which grows with the horizon and stops at the unconditional spread 1.72. The dashed pair is the same band computed at the true parameters, which nobody has. The marks past zero are what actually arrived: 12 of 14 inside the band this once, which is one draw and settles nothing.

What the model says next

The usual account of a time series stops at estimation. A forecast asks the other question — not what the parameter is but what the next observation will be — and the band round it is a closed form that grows with the horizon and then stops growing, at a value the series was going to reach anyway.

9 figures · Forecast, part 1 What a 95% forecast interval covers, counted. 1200 series of 25 observations from an AR(1) with φ = 0.7, at each horizon, on one set of seeds. The upper line is the interval computed at the true parameters — it covers 95.3% on average, which is the check that σ²Σψ² is the right formula rather than a claim about anything a forecaster can do. The lower line is the same formula fed σ̂² and φ̂: 92.8% at one step and 87.3% at 6. The interval that would cover what it claims is 6.9% wider at one step.

The interval that forgets it estimated

The forecast band is derived for a model whose parameters are known, and then computed by putting estimates into it. Counted, the 95% interval covers 87.3% six steps ahead on twenty-five observations, and the point forecast inside it returns to the mean a third faster than the series does.

9 figures · Forecast, part 2 What each criterion selects, at 50 observations. 700 series from an AR(2) with coefficients 0.6 and -0.3, every order from 0 to 8 fitted to the same 42 responses so the log-likelihoods are comparable. AIC finds the true order 55.1% of the time and lands above it 25.7%; BIC finds it 54.1% and lands above it 4.0%. The closed form for one extra lag is P(χ²₁ > 2) = 15.73% for AIC, which does not depend on n at all, and P(χ²₁ > ln n) = 4.79% for BIC at this size, which falls to zero. Under the true order is the other failure and it is BIC's: 41.9% against 19.1%.

Choosing the order

One criterion is consistent and one is not, which is the whole of what gets said about them. At two hundred observations the consistent one is right 95% of the time and the other 70%; at fifty they are both right 54% of the time and wrong in opposite directions, and consistency has not started to mean anything yet.

9 figures · Order selection, part 1 What the interval covers once the order is chosen as well. 1200 series of 40 observations from an AR(1) with φ = 0.7, at each horizon, on one set of seeds. The upper line is the interval computed at the true parameters — it covers 95.5% on average, which is the check that σ²Σψ² is the right formula rather than a claim about anything a forecaster can do. The lower line is the same formula fed σ̂² and φ̂: 93.6% at one step and 90.8% at 6. The third line chooses the order by AIC from the same data before computing the interval, which costs a further 0.8 points at h = 6.

The interval after the choice

Estimating the coefficients of a known model costs a 95% forecast interval about two points of coverage. Choosing which coefficients to estimate, from the same forty observations, costs another four and a half — so the step nobody records in the output is the more expensive of the two.

9 figures · Order selection, part 2 Which side the misses fall on, skewed shocks, 95.0% intervals. With skewed shocks and the model's parameters known, the share of 10,000 forecasts that fall below and above a nominal 95.0% normal interval, at each horizon: 1 ahead, 0.0% below and 5.2% above; 2 ahead, 0.0% below and 4.9% above; 4 ahead, 0.0% below and 4.6% above; 8 ahead, 0.0% below and 4.2% above; 12 ahead, 0.0% below and 4.9% above.

Shocks that are not normal

Every forecast interval in this field reads a normal quantile, and every series it was scored on had normal shocks. Give a first-order autoregression shocks with heavy tails and the 95% interval still covers about 95%; its 99% interval covers 98.2%. Give it skewed shocks and the 95% interval covers 94.8% one step ahead with every miss on one side — 0.0% below, 5.2% above — and the horizon does not repair it: twelve steps ahead the misses are still 0.0% below and 4.9% above, because a persistence of 0.7 keeps the error's skewness at 1.11 however far out the forecast goes. A two-sided coverage count cannot see any of this.

3 figures · Forecast, part 3 Which side the misses fall on. Misses of a nominal 95% forecast interval 1 step ahead on 60 observations of a first-order autoregression at 0.7, split by side. normal shocks: the interval as printed 3.1% below and 3.1% above, shocks from the residuals 3.8% below and 4.2% above, the same, coefficient re-estimated 3.3% below and 4.4% above. Student's t on five: the interval as printed 3.3% below and 3.0% above, shocks from the residuals 3.9% below and 3.0% above, the same, coefficient re-estimated 3.6% below and 3.1% above. skewed shocks: the interval as printed 0.0% below and 6.5% above, shocks from the residuals 4.3% below and 4.1% above, the same, coefficient re-estimated 1.8% below and 4.3% above.

A tail sixty residuals have barely seen

A forecast interval that reads a normal quantile has every one of its misses on one side when the shocks are skewed, and the natural repair is to draw future shocks from the model's own residuals. It works on the side: 0.0% below and 6.5% above becomes 4.3% and 4.1%. It pays for that with coverage, 93.5% to 91.6%, and on sixty observations it does this under normal shocks too. And a nominal 99% interval has a ceiling nothing in the resampling can lift, because one new shock beats all of sixty old ones in either direction with probability 3.3%.

3 figures · Forecast, part 4

The criterion, and what it assumes

D-optimality gave the catalogue back, which was reassuring, and left two things unsaid. A, D and E are one family and the letter is a choice — the design that wins the first is nearly worst at the last. And for a model whose information depends on its own parameters, a design is optimal only at a guess about the answer, which makes the cost of guessing wrong a quantity with a closed form.

The end of the family is the member the algorithm cannot reach. Every member of the Φₚ family optimised on the same 11² candidates, and two eigenvalues of each answer. The upper curve is the smallest eigenvalue of the information matrix — the quantity E-optimality maximises — which rises from 0.0993 at the D end to 0.1994 at p = 64. The lower curve is the gap between that eigenvalue and the next one up, which falls from 0.0611 to 0.0011. A smallest eigenvalue is not differentiable where it is repeated, and the family is driving the gap to zero: the one criterion here whose meaning fits in a sentence is the one whose optimum sits on a corner of its own surface. The candidate grid is on a slider and it answers a narrower question than it looks. On this square region, refining an odd grid from seven to eleven moves nothing at all — the optimum's support is the corners, the edge midpoints and the centre, and every odd grid from five up contains all of them. An even grid has no centre point and cannot reach the answer at any member of the family. On a disc, where the boundary passes through no grid point, refinement does move it.

The family behind the letters

A, D and E are not three ideas. They are three points of one family with a single dial, and running the dial from one end to the other doubles the smallest eigenvalue of the information matrix while closing the gap above it fifty-three-fold — which is the family driving its own last member to the place where it stops being differentiable.

9 figures · Criterion, part 4 Where a Ds-optimal design puts its runs. The Ds-optimal measure over 121 candidate settings on a square region. It keeps 9 of them and discards the rest, and the 9 it keeps are the settings a catalogue would have offered without any of this arithmetic. What the search adds is the weights: 0.2500, 0.1250, 0.0625, which nine equal runs cannot express.

The two terms anybody wanted

D-optimality estimates all six parameters of a quadratic as precisely as possible. Nobody wants that. An experimenter looking for a maximum wants the two curvature terms, and the design that gives them is not the D-optimal one — it is a quarter of the runs at the centre, exactly, and the D-optimal design is 75.3% efficient for the question that was actually asked.

8 figures · Criterion, part 5 Where to look depends on the answer. The information a single run at time t carries about the rate of an exponential decay, (∂η/∂θ)² = t²·exp(−2θt), at three values of θ. Each curve has one maximum and it is at t = 1/θ exactly — marked, and found by a search over 8,001 settings that was never told the formula. Nothing in a linear model behaves this way: there the information matrix is X′X and the parameters are not in it, so a design can be chosen once and used whatever the answer turns out to be. Here the design is optimal at a guess, and the three curves are three different experiments for one model.

The design that needs the answer

Every design this site has computed is optimal whatever the experiment turns out to say, because X′X does not contain the parameters. For a non-linear model it does, so the best place to take a measurement is a function of the number the measurement exists to find — and guessing it three times too low costs two and a half times more than guessing it three times too high.

9 figures · Local design, part 1 A design that is right once, against one that is never wrong by much. Two designs for the same two-parameter model, scored at every true value of K across a range of 16-fold. The local design is the two-point optimum for a guess of K = 1: it reaches 100% there and 66.7% at the worst point of the range. The hedged design maximises the average of log|M| over a prior spanning a factor of 4 either side, uses 3 settings rather than two, and is never below 75.4%. What it costs is 10.4 points at the one value the local design was built for — which is the whole trade, and it is only available to somebody willing to say how wrong the guess might be.

The design that hedges

A locally optimal design is right at one value of the unknown and 23.9% efficient at the edge of a sixteenfold range. Averaging the criterion over a prior instead buys the worst case back to 56.3% — and buys it by adding support points, at spreads the arithmetic decides rather than the experimenter — a third setting at a factor of 3.36 and a fourth at 8.86.

8 figures · Local design, part 2 What the main effects cost the rest of the model. The compound optimum of α·log Ds(main effects) + (1 − α)·log D on an 11-by-11 grid of the square, at 14 weights from α = 0 to 0.999, placed by its D-efficiency against the D-optimal design and its main-effect efficiency against the factorial. The D-optimal design (α = 0) gives the main effects 74.3% of what the factorial gives. At α = 0.5 the design keeps 94.6% of D and gives 85.6%; at 0.9, 64.9% and 96.8%; at 0.99, 31.3% and 99.7%. The factorial itself (open, at the left edge) gives the main effects everything and the model nothing: its D-efficiency is zero. The design optimal for the two quadratic terms (red) gives the main effects 50.0% at a D-efficiency of 83.6%.

A subset the model cannot use

Nominate the two main effects of a quadratic response surface as the parameters of interest and the optimal design is the 2² factorial — four corners, a quarter each — which leaves the full model's information matrix at rank four of six and the curvature with nothing at all. That is why the field's algorithm would not converge on it. Approached through a compound criterion instead, the trade is explicit: 96.8% of the main effects' best information costs the model 35% of its D-efficiency, and the last three points cost nine tenths of what is left about the curvature.

5 figures · Criterion, part 6 What a stated power leaves of the main effects. The main-effect efficiency of the best design reaching a power of 0.8 at level 0.05, as N|δ|² rises through 50, 60, 70, 80, 100, 130, 170, 220, 300, 400, 600, 1000. a saddle, two-degree test, the compound path: out of reach, out of reach, 0.806, 0.841, 0.881, 0.914, 0.937, 0.952, 0.966, 0.975, 0.984, 0.990. a bowl, two-degree test, the compound path: 0.782, 0.823, 0.851, 0.871, 0.898, 0.923, 0.941, 0.955, 0.967, 0.976, 0.984, 0.990. a bowl, its own one-degree contrast, the compound path: 0.827, 0.859, 0.880, 0.896, 0.918, 0.937, 0.952, 0.964, 0.973, 0.980, 0.987, 0.992. a bowl, a factorial with centre points: 0.914, 0.930, 0.940, 0.948, 0.959, 0.969, 0.976, 0.982, 0.987, 0.990, 0.993, 0.996.

A design chosen for a test

A screening experiment that wants its main effects and a check on curvature has been told to trade them with a weight nobody can state. It can state a power. Read against a power of 80%, the frontier collapses to one curve in runs times the squared size of the curvature, and a large enough trial pays almost nothing. But the curve belongs to a test. A factorial with a few centre points beats the entire frontier on a bowl, spending 4% of its main-effect information where the frontier spends 10%, and against a saddle of the same size it has exactly its 5% level.

3 figures · Criterion, part 7

Balancing on what was recorded first

The adaptive field's rules read outcomes and every one of them broke something. These read baseline covariates, and each holds one imbalance flat while the others grow behind it exactly as a coin's do. What follows is the opposite of the outcome-adaptive story at every step: the unadjusted analysis is too cautious rather than too eager, and the exact test that cost nineteen points of power there costs almost nothing here.

What each rule leaves behind, at 120 patients. Four allocation rules over the same cohorts and the same seeds, each scored on three imbalances: the number of patients in each arm, the worst of the nine factor levels, and the worst of the 24 cells of the cross-classification. No rule holds all three. Permuted blocks hold the totals exactly and leave the margins near a coin's. Blocks inside every cell hold the cells and let the totals drift, because 24 part-filled blocks do not have to end level. Minimisation holds the margins and the totals and is at 83% of a coin's cell imbalance. Each of the three columns is somebody's definition of a balanced trial.

Balancing what is known in advance

Four allocation rules, three definitions of balance, and no rule that holds more than one of them. Minimisation keeps the worst factor margin near three patients whether the trial has forty or six hundred and forty — and lets the imbalance in the cross-classified cells climb to 86% of a coin's, because the cells are not what it is watching.

9 figures · Assignment, part 1 What a rule gives away by being predictable. Minimisation run at every probability from a coin to fully deterministic, over 500 cohorts of 120 at each. The upper line is the share of assignments an investigator who knows the rule and the enrolled patients can name in advance: 49.7% at p = 0.5, which is a coin and cannot be beaten, and 87.6% at p = 1 — short of everything only where the two arms tie and the rule falls back on a coin. The lower line is what that is worth: an investigator who enrols a patient 0.5 of a standard deviation better than average whenever they predict their favoured arm produces a treatment effect of 0.75 where the truth is zero. Nothing about the randomisation was broken; the bias entered through who was enrolled, which is the one thing an allocation rule cannot control. The dashed line is the closed form 2δ(2g − 1).

The rule that can be guessed

A balancing rule improves as it becomes more deterministic, and a deterministic rule can be worked out in advance from information the person enrolling the patient already has. At full determinism 87.6% of assignments are guessable, and an investigator who acts on the guess produces a treatment effect of three quarters of a standard deviation where the truth is zero.

9 figures · Assignment, part 2 Four analyses of the same trials, with no treatment effect at all. 700 trials of 120 patients allocated by minimisation at p = 0.8, with the prognostic factors carrying a real effect on the outcome and no treatment effect — every rejection below is a false one. Two statistics, the plain difference and the same after adjusting for the balanced factors, each read against two reference distributions: a t table, and the set of allocations the rule could have produced from these covariates. The unadjusted comparison rejects 0.6% where it claims 5% — conservative, which is a loss of power rather than an error, and nothing on the output says so. Adjusting puts it back at 5.4%. Both re-randomised versions are at their nominal level by construction, whatever statistic goes into them.

The analysis has to know the rule

A trial balanced by minimisation and analysed by comparing the two arms' means rejects a true null 0.6% of the time where it claims 5%, and at full determinism 0.0%. That is not an error anybody complains about — it is a test that has stopped working, paid for by a balance the analysis then refused to use.

9 figures · Assignment, part 3 The allocations this trial could have made, and the ones it could not. One 120-patient trial allocated by minimisation at p = 1, re-randomised 399 times. No outcome is redrawn anywhere in this figure: each re-randomisation runs the rule again over the same patients in the same order with the same recorded factors, so what is drawn is the set of experiments that could have happened. The bars are that set; the outline is what shuffling the labels gives, which is the reference distribution of a coin and is what every off-the-shelf permutation routine assumes. The coin's is wider — its 5% point is 1.95 against the rule's 1.09 — because a coin's allocations are less balanced and a less balanced allocation gives a larger statistic. Reading this trial against it makes the test conservative rather than anti-conservative, which is the opposite error from the outcome-adaptive case and for the same structural reason.

The reference the covariates supply

Hold the outcomes fixed, re-run the rule that assigned them, count. The same construction cost nineteen points of power in the adaptive field, because its rule chased outcomes and its critical value depended on a rate nobody has. Here the rule reads only what was recorded before anything happened, and the same unadjusted statistic goes from 20.3% power to 55.0% by being read against the right distribution.

9 figures · Assignment, part 4 Four intervals for the effect, minimisation at p = 0.8, 120 patients. Under minimisation at p = 0.8, 120 patients, over 600 trials with an additive effect of 0.45 and 199 re-run allocations per trial, the average width of four 95% intervals for the effect, and how often each covers it: unadjusted t interval 1.229 wide, covering 99.7%; unadjusted, test inverted 0.816 wide, covering 95.3%; adjusted t interval 0.723 wide, covering 96.0%; adjusted, test inverted 0.722 wide, covering 96.2%.

The interval the test inverts to

A randomisation test of no effect becomes an interval for the effect by testing every value at once: subtract the hypothesised effect from the treated outcomes and ask the same question again. Because the rule reads only covariates, one set of re-run allocations answers every value. Under minimisation the unadjusted t interval covers 99.7% at a width of 1.229; the inverted test of the same unadjusted difference covers 95.3% at 0.816, a third narrower, with no model of the outcome. Under complete randomisation the two are the same interval — which is what says the third was the rule's.

4 figures · Assignment, part 5

FieldsThreadsSeriesConceptsFigure librarySearch