The likelihood ratio for the two binomial probabilities is
strictly increasing in . By the Neyman-Pearson lemma, reject in the highest upper tail. Under the null,
Hence reject exactly when , with no randomization required. Its statistical power at the stated alternative is
A possibly randomized statistical test has rejection probability after observing . Its statistical power at parameter is , and its test size is the supremum of that statistical power over the null parameter set.
For simple hypotheses with densities , the Neyman-Pearson lemma states that a level- test is most powerful if it rejects where , accepts where , and randomizes on equality to obtain null rejection probability . Here is chosen to reach that test size. To prove optimality, let be this test and any test of level at most . Pointwise,
Integrating and using gives . This proves maximal statistical power. For , equality also forces any equally powerful test to use all its allowed test size and agree with the indicated decisions away from the likelihood-ratio tie set, up to null sets.
For the Poisson sample, and, for every fixed , the likelihood ratio is
It increases strictly with . Choose an integer so that , and reject for , with boundary rejection probability
This has exact test size . Its rejection ordering and calibration do not depend on which is selected, so the lemma proves it uniformly most powerful for the whole one-sided alternative. Randomization is generally needed for an exact arbitrary test size in a discrete model.
For the approximate design, the central limit theorem under gives critical sum . Under , the sum is approximately normal with mean and variance . Target statistical power therefore gives
or . Thus the requested design expression is
Round upward to obtain an integer design. This is a normal approximation to the requested statistical power, rather than an exact discrete sample-size guarantee; the exact Poisson test above provides its exact calibration.
The significance level of a test is a bound on its probability of rejecting a true null; for a composite null the bound is uniform over its parameter values. Its power at an alternative parameter is its rejection probability under that parameter.
The Neyman-Pearson lemma concerns a simple null density and a simple alternative density for the whole sample. Among tests of size at most , a most powerful test rejects where , accepts where , and if necessary randomizes on equality so that its size is . Let its rejection probability function be , with , and let be any competing test of size at most .
Pointwise,
on the rejection side both factors are nonnegative, and on the acceptance side both are nonpositive. Integrating gives
Thus the likelihood-ratio test has at least as much power as any competitor. Even when the sample has a continuous density, the likelihood ratio can have an atom, so boundary randomization may still be necessary in the general statement. For the exponential test below the statistic has a continuous distribution and no randomization is needed.
For a fixed alternative , the joint likelihood ratio is
It is strictly increasing in . By the Neyman-Pearson lemma, the most powerful level- rejection region is therefore . Under the null, with the second parameter interpreted as a rate. Its continuous distribution gives
This threshold depends on the null and , but not on the particular smaller alternative . Thus the same test is most powerful for every , making it uniformly most powerful.
Under a true alternative , , so its attained power is
For each , the Neyman-Pearson lemma bounds the power of every other level- test by this value. Hence the displayed function is not only an upper bound but the attainable power envelope on the specified alternative.
The log likelihood ratio statistic comparing the unrestricted fit with the independence fit is
The unrestricted model has free parameters. Independence has , so the difference is . By Wilks theorem, under the null hypothesis the statistic has asymptotically a chi-squared distribution with degrees of freedom, rather than , , or .
For the supplied statistic, the approximate p-value is . Thus reject independence at the 5% level, but not at the 1% level: the data provide evidence of an association. The approximation is reasonable here; the smallest fitted cell count is , so all fitted counts exceed . The statistical hypothesis test concerns association, not a causal effect of personality.
The likelihood ratio between the two normal means is
Since , it is strictly increasing in . The Neyman–Pearson test therefore rejects for large sample means. Under the null, , so choosing a right-tail probability gives
This has size exactly because the sampling distribution is continuous.
For every particular alternative mean , the likelihood ratio is increasing in the same statistic . Its size- threshold is fixed solely by the null distribution, so it is always the same found in part (c), independent of .
The single rejection region is therefore most powerful against every member of the composite alternative. Hence the test is uniformly most powerful for .
For , the likelihood ratio in favor of the larger variance is
Because , this is strictly increasing in . The Neyman-Pearson lemma therefore selects an upper-tail rejection region. Under the null, , so size is obtained with
Continuity of the chi-squared distribution makes the boundary have probability zero. This is a fixed-mean normal variance test.
The raven paradox combines a plausible equivalence principle with an overly unrestricted account of positive-instance scientific confirmation. Let and mean that is a raven and black respectively. Then
A white shoe is a positive instance of the right-hand generalization. If every positive instance confirms its universal generalization, and logically equivalent statements receive the same evidential support, inspecting the shoe seems to confirm the original raven claim. There is no contradiction in the equivalence itself. The tension concerns the evidential rule and the intuition that an arbitrary shoe teaches us nothing about birds.[ref]
The right response is to make background knowledge and sampling explicit. Under Bayesian confirmation, evidence increases support for precisely when its likelihood is greater under than under the alternative, assuming nondegenerate prior probabilities. A fact can therefore be compatible with yet provide no new support. The mere existence of a white shoe does not determine the needed likelihood ratio.[ref]
Consider a deliberately simple finite model. There are objects, exactly ravens, and nonravens, with . All nonravens are white. Under all ravens are black; under exactly one raven is white. Draw uniformly from the entire population. The event of drawing a white nonraven has probability under either hypothesis. Its Bayes factor is one and it leaves the prior odds unchanged. A black raven instead has likelihoods and , so gives Bayes factor for .
Now change the protocol: draw uniformly from the white objects, then discover that the selected object is a nonraven. This result has likelihoods and . It gives the small positive Bayes factor
For and , this is , whereas sampling a black raven from the full population gives . The white nonraven can now help because the search could have revealed a white raven. The two experiments have different selection information, so their differing evidential values do not violate logical equivalence.
This model establishes possibilities, not a universal numerical law about shoes. Different alternatives, population information or selection procedures change the result. If the shoe's being a nonraven was already known, discovering that it is white need not add anything. Conversely, ruling out a potential counterexample can supply a small amount of information. Retain logical equivalence, but reject the unconditional instance-confirmation rule. Once the complete evidential situation is specified, neither zero support nor weak positive support is paradoxical, and neither amounts to proof of a universal law.
Let mean that is a raven and that it is black. The hypothesis
is logically equivalent to . If the observed paper is both nonblack and a nonraven, it is a positive instance of the second formulation. Combining an unrestricted positive-instance rule with preservation of scientific confirmation under logical equivalence produces the raven paradox.[ref]
Whether the observation actually supplies evidence depends on what was already known and how the object was selected. In Bayesian confirmation, with background , support means . If the observer already knows the selected item is a nonraven and its colour is conditionally independent of , then : the likelihood ratio is one and there is no confirmation. If all of was already known, learning it again also gives no update.
Under a different sampling scheme, the same classification can provide slight support. In the finite-population model of raven confirmation, fix ravens and nonblack objects. Compare no nonblack raven with exactly one. Sampling uniformly among the nonblack objects gives a nonraven with probabilities and , hence likelihood ratio . If is large this is tiny evidence, not a proof. This example establishes possibility under explicit assumptions, not that every white-paper observation has that evidential role.
The appropriate answer is therefore conditional: mere logical instantiation does not settle the support supplied by observing this paper. State the background and sampling procedure. A familiar paper deliberately selected as a known nonraven normally supplies no new evidence about raven colour; a specified alternative sampling model may yield a small positive update.
A Bayesian confirmation account can dissolve the apparent conflict between logical equivalence and ordinary judgements about relevant evidence. It does so by rejecting an unrestricted positive-instance rule and specifying how the evidence was sampled. Bayes' theorem alone does not guarantee that every white shoe is irrelevant, or that every black raven provides strong support.
Let assert that every raven is black. Its contrapositive is logically equivalent to , so the two descriptions must receive the same conditional probability. The raven paradox arises if every positive instance of a universal statement automatically confirms it: a white nonraven then confirms the contrapositive and hence . The disputed step is that unrestricted confirmation rule, not the logical equivalence.[ref]
For background , set , and . For and positive evidence probability,
Thus it is the comparison of likelihoods, including the sampling protocol and alternatives, that determines support. Merely rewriting cannot change it.
The finite-population model of raven confirmation makes the intuitive difference quantitative. Fix ravens and nonblack objects. Compare an all-black-ravens hypothesis with the alternative that exactly one raven is nonblack. A uniform draw among nonblack objects that finds a nonraven has likelihood ratio . A uniform draw among ravens that finds a black one has likelihood ratio . If , the former is much nearer one. With equal prior probabilities, and give posterior probabilities approximately and . A tiny amount of confirmation need have no practical significance.
This example is not a theorem about every possible background. Inspecting an object already known to be a white shoe adds no relevant information about its being a nonraven. Other alternatives or selection mechanisms change the likelihoods and can remove or reverse the support. Conditioning only on the observed colour while silently changing how the object was selected would compare different experiments.[ref]
The resolution is therefore qualified. A coherent probability model preserves equivalence and explains why superficially parallel observations need carry radically different evidence. It does not select a unique prior or sampling model without scientific judgement. Those remaining modelling questions are limitations of an application, rather than a contradiction in Bayesian inference.
Scientific underdetermination limits what evidence alone establishes, but its practical consequences depend on the strength of the equivalence being claimed. Rival theories fitting present data need not agree on future observations. Rival formulations agreeing on every observable in a specified domain present a more persistent limitation within that domain.
With finitely many data points , suppose an interpolation polynomial fits all measurements. Then
also fits them for every . This proves nonuniqueness for a deliberately broad model class. It does not prove that all these functions are equally plausible physical theories. Constraints from independently tested mechanisms, bounds on derivatives, symmetries and the measurement process can sharply restrict the class. A new observation at a different point can also distinguish the displayed alternatives.
The Duhem problem concerns an additional kind of latitude: a prediction tests a theory together with auxiliary hypotheses. A mismatch can be assigned to different components. This is a reason to vary apparatus, calibrate independently and use multiple kinds of evidence. It is not a demonstration that any preferred theory can be made equally well supported by arbitrary repairs.[ref]
Exact empirical equivalence requires more. If two accounts supply identical probability distributions for all observations allowed by a stated setup, observations of that kind cannot favour one through their likelihoods. Bayesian inference makes the limitation explicit: posterior odds equal prior odds when the likelihood ratio is one. Declaring a prior preference does not transform it into new experimental discrimination.
Theories can nevertheless differ in explanatory organization, computational usefulness or how they connect with a larger domain. Those are legitimate questions, provided their role is stated. If a proposed new intervention distinguishes them, the equivalence did not cover that enlarged domain. If the difference is instead only a change of redundant description, treating it as two competing physical worlds may create a spurious dispute; gauge symmetry provides an important case requiring care.
For scientific practice, the appropriate response is model comparison with explicit assumptions, searches for discriminating tests and honest limits on what has been established. For scientific realism, enduring empirical equivalence challenges any claim that the evidence uniquely fixes a full ontology. It does not make precise predictions or experimental decisions impossible. Underdetermination is consequently a constraint on inference whose seriousness must be demonstrated case by case, rather than a general proof that all scientific reasoning is arbitrary.
Scientific confirmation is evidential support rather than deductive proof. Two distinct problems concern, first, what counts as relevant positive evidence and, second, which part of a tested theoretical package that evidence supports. A defensible response makes background information and the production of evidence explicit. The following two problems are independent: resolving the logical puzzle about positive instances does not eliminate ambiguity about auxiliary hypotheses.
Problem one: the raven paradox. Write the hypothesis as
where means “is a raven” and “is black”. Its contrapositive is
The hypotheses are logically equivalent. A plausible equivalence principle says that identical information must support them equally. A plausible instance principle says that an observed that is confirms “every is ”. Combining the principles makes a nonblack nonraven, such as a white shoe, support the raven hypothesis. That conflicts with the intuition that inspecting shoes tells us nothing about ravens. Hempel develops the puzzle in his analysis of instance-based scientific confirmation.[ref]
My preferred response retains logical equivalence while rejecting an unconditional instance rule. By Bayes' theorem, for evidence , background and positive relevant probabilities,
An instance's bearing depends on how likely it was under the competing possibilities, including how the object was selected. No rule depending only on the grammatical form of an instance can settle that comparison. Branden Fitelson and James Hawthorne analyse the role of background assumptions and Bayesian confirmation in this puzzle.[ref]
Here is an explicit illustrative model. There are 100 labelled objects: ten known ravens and ninety known nonravens. Under all ten ravens are black; under the sole alternative , exactly one of the ten is nonblack, with its identity uniformly distributed. Nonraven colours are already known and identical under both possibilities, and the prior probabilities of are equal. Selecting a known white shoe according to a rule independent of has the same likelihood under each, so the posterior probability of stays . A uniformly selected raven, however, is black with probability under and under . After observing it to be black,
If sampling is instead from known nonblack objects whose membership is uncertain, discovering a nonraven can rule out a possible counterexample and provide support. There is no context-independent prohibition on support from a nonraven. The response explains both the familiar irrelevance intuition and why it is not universally correct. Its limit is that the prior probabilities and sampling assumptions need independent defence; the calculation is a transparent model, not a theorem that all shoes are irrelevant.
Problem two: the Duhem problem of testing a package. Suppose the target hypothesis and auxiliary hypotheses jointly imply an outcome :
A failure gives , equivalently , rather than alone. Success also does not establish either premise: that would affirm the consequent. Pierre Duhem's point is especially pressing for physical experiments whose predictions require theories of the apparatus as well as the target system.[ref]
My preferred response combines independent calibration, deliberately varied tests and explicit comparison of packages. For a simple model, suppose a detector has independently calibrated false-positive probability and detection efficiency . A target predicts a signal on each trial with probability . Then the positive-output probability is
Under a no-signal rival it is . If , , and , then . One positive reading supplies a Bayes factor of for the specified target package against the specified no-signal package. But an uncalibrated faulty detector with false-positive probability fits the same positive-reading rate. Target data alone do not distinguish those packages.
Signal-absent calibration trials test the false-positive assumption directly. If ten independent such trials are all negative, their likelihood ratio in favour of against is
This does not magically isolate : the calibration relies on its own assumptions, and correlated trials would invalidate the power calculation. It shows concretely how additional evidence can discriminate explanations that the initial data leave unresolved. Repeating the target measurement with apparatus based on different physical processes further reduces dependence on one shared failure mode.
Thus the Duhem problem defeats automatic attribution of support or failure, not rational experimental comparison. Where rivals remain empirically equivalent, suspension of judgement about their difference is appropriate. Together, the two responses defend scientific confirmation as a structured, fallible comparison using background information, rather than either pure instance counting or a mechanical verdict on an isolated hypothesis.
The statistical power of a test is , its rejection probability at parameter . Its size is ; for a simple null this is just .
The Neyman-Pearson lemma states that for testing a simple null probability density function against a simple alternative probability density function , a test rejecting where , accepting where it is below , and if needed randomizing on equality to attain the desired size is most powerful among all tests of that size. It follows by integrating and using the size constraint.
The proposed probability density function is nonnegative, and
For an independent sample, cancellation of the factors in the likelihood ratio gives
For this is increasing in . The most powerful test therefore has rejection region
The sum has a continuous positive probability density function on , so no boundary randomization is required. The threshold depends on the null and chosen size, but not on which was specified. The same test is therefore a uniformly most powerful test for the composite alternative .
The multinomial likelihood is a parameter-independent factor times . Maximize its logarithm subject to : the multiplier equations give , and summation makes . Thus . Positivity of all counts makes this the unique interior maximum by strict concavity.
The likelihood ratio compares the fitted null likelihood with this saturated likelihood. Taking minus twice its logarithm gives . Under regularity and large expected counts, compare it to for one fitted parameter: the saturated probabilities have free coordinates and the null has one. If the parameter is fixed at , use and . Reject for a value above the chosen upper quantile.
For the ungrouped logarithmic-series observations write and . The log likelihood is
Setting its derivative to zero gives
Bisection of this monotone mean equation, or Newton's method with a valid bracket, obtains .
The supplied table instead uses ten grouped cells, with the last probability . Its likelihood is determined by the cell counts and is not the ungrouped likelihood just differentiated. From its listed values, the minimum deviance occurs near ; direct minimization of this grouped deviance gives about . The supplied two forms of the estimate therefore concern different likelihoods. They must not be identified: the full-data score equation gives about , while the grouped table points near . The following requested tests use the table's grouped likelihood, with ten cells. The source values and totals are preserved unchanged in TeX.
For a fixed , the likelihood ratio against the null is
strictly increasing in . The Neyman–Pearson critical region is therefore an upper tail in , and its null threshold depends only on the desired size, not on which was chosen. Thus the same test is uniformly most powerful for the entire one-sided alternative.
Under the null with the second parameter interpreted as rate. Its continuous distribution permits the exact threshold
Under any fixed alternative its rejection probability is . Applying the lemma separately to each such shows that every level- test has power at most this value there. The constructed test attains the bound for all simultaneously.
The generalized likelihood ratio compares the constrained maximum from part (b) with the unrestricted maximum from part (a). Taking minus twice its logarithm gives
The unrestricted model has free parameters; the independence model has . Thus the asymptotic null distribution is , with four degrees of freedom. The supplied value gives
Therefore reject independence at the 5% level, but not at the 1% level. There is evidence of an association between the two classifications. The fitted expected counts are all about nine or larger, so the usual chi-squared approximation has no very small expected cells here.
Let denote the standard normal density. Under the shifted sampling density , the likelihood ratio is . Simulate independent and use
Multiplying the weight by restores the original density, so each summand has expectation . The estimator is unbiased. For the second moment,
because . Since ,
For , and , so , proving . Nonnegativity of variance also implies ; division by the positive gives
Dispositional attribution is most biased when a person's conduct is easy to encode as a trait but the constraints on that conduct are poorly represented or insufficiently used. The bias is conditional on attention, cognitive load, expectations and learned explanatory practices. Calling an inference biased requires a defensible benchmark: behavior sometimes really is informative about a person's enduring preferences. Correspondence bias concerns excessive correspondence between observed conduct and inferred disposition after relevant situational causes are taken into account, rather than the mere existence of a trait judgment.
The classic demonstration separates the direction of a political essay from whether its author chose that position. In the assigned-position condition, readers nevertheless inferred attitudes corresponding to the essay. Readers also distinguished choice from assignment, so the result supports insufficient discounting rather than complete blindness to constraint. The experiment by Edward E. Jones and colleagues makes the situation explicit and therefore challenges the explanation that perceivers simply had no situational information. It does not show that all information in an assigned essay is necessarily nondiagnostic: enthusiasm, wording or willingness to comply could still be informative under a different task model.[ref]
Conditional diagnosticity in attribution provides a useful normative benchmark. Let denote a disposition, the observed behavior and the situation. By Bayes' theorem, posterior odds equal prior odds times a conditional likelihood ratio:
If assignment makes the same behavior almost equally likely under either disposition, its likelihood ratio is near one. Treating this behavior as if it had been freely chosen exaggerates its diagnosticity. Conversely, repeated voluntary actions across varied situations can justify a strong dispositional attribution. The benchmark must therefore distinguish an inference from a single constrained act from an inference supported by a history of choices.
One mechanism is asymmetric attention. The actor is a discrete, changing object; the constraints can be dispersed across instructions, other people and background circumstances. A perceiver may accurately remember that a constraint existed without representing its causal force when interpreting the action. Explaining a nervous performance as anxiety is also economical: one familiar trait compresses many movements, whereas reconstructing an unfamiliar interview situation requires additional assumptions. This reasoning predicts more correspondence bias when the actor dominates attention, contextual information is ambiguous or perceivers lack relevant situational knowledge. Social salience alone, however, does not identify which trait will be inferred; expectations and category knowledge also shape interpretation.
The cognitive-load account receives a more direct test in Daniel T. Gilbert and colleagues' experiments. Participants watched silent videotaped clips of an apparently anxious target, with subtitles describing her discussion topics. Those required to rehearse the topics in order used this situational information less to adjust their trait judgments. The load therefore involved processing the context itself, rather than simply removing that information. Related experiments used constrained speech. The proposed sequence distinguishes identifying the behavior, deriving a characterization and correcting for context; the last step was especially vulnerable to cognitive load. This explains why merely telling observers about a constraint need not remove correspondence bias.[ref]
The experimental manipulation supports a contribution from limited cognitive resources, but it does not prove that every social attribution follows a rigid three-stage sequence. A concurrent task may also alter attention to instructions or memory for diagnostic details. Context could influence the earliest interpretation when it is obvious and familiar. Thus the strongest claim is that effective situational use can require additional processing in these tasks, rather than that dispositional inference is universally automatic and situational inference universally controlled. A discriminating experiment would measure comprehension of the constraint as well as the final social attribution and vary processing time independently of contextual clarity.
Culture also changes the starting point of explanation. Michael W. Morris and colleagues compared American and Chinese explanations of social and physical events, using animated displays and accounts of killings in newspapers. The social explanations differed in their emphasis on individual characteristics and contextual causes; the comparison with physical events constrained an account based on a general inability to reason about causes. These findings suggest that learned explanatory frameworks influence which causes come readily to mind.[ref]
These comparisons do not warrant treating national populations as uniform psychological types. Nationality was not randomly assigned, newspaper writing is not a transparent measure of private judgment, and language or knowledge of the events can affect the explanations elicited. Their theoretical importance is that a supposedly universal default must accommodate systematic variation. An observer who routinely represents relationships and constraints may need less later correction, rather than merely possessing more cognitive resources. Cross-cultural psychology therefore complements the resource account by asking what information is represented before correction becomes necessary.
Motivation matters through the same conditional logic. A consequential decision can encourage searching for alternatives, but a desire for a quick, blame-consistent explanation can instead favor a dispositional attribution. Neither effect follows from motivation alone; the relevant question is which information the motive makes useful. Clear constraints, sufficiently understood instructions, time to use them and observations across multiple settings should improve calibration. These are mechanistic predictions, not guarantees that deliberation removes every cognitive bias.
The evidence is best explained by an interaction between behavioral diagnosticity and the perceiver's representation of context. Correspondence bias is strongest when salient conduct is readily characterized but situational alternatives receive too little weight; it weakens when constraints are understood and effectively incorporated. This account preserves the genuine usefulness of trait inference while explaining why its accuracy varies across tasks, resources and cultural experience.
For fixed , the likelihood ratio is strictly decreasing as increases. A likelihood-ratio test rejects for small , so its rejection rule is equivalently large . Under the null hypothesis, the standard normal distribution and chi-squared distribution results give independent quantities
Their ratio therefore has Student's t-distribution with statistical degrees of freedom. The exact level- rule is
This is precisely the two-sided Student's t-test; choosing the corresponding cutoff for yields the same rejection region and the same p-value.