Study guide: Combinatorics and Diagnostic Testing

Course Outline

  1. Solving Statistical Exercises
  2. Descriptive Statistics and Survey Data
  3. Counting Models and Combinatorial Principles
  4. Six Health-Service Counting Scenarios
  5. Contraintes de dénombrement
  6. Probabilités conditionnelles et Bayes
  7. Mesures d’un test diagnostique
  8. Dépistage des boîtes falsifiées
  9. Diagnostic Test Sensitivity and Specificity
  10. Recognizing Discrete Probability Laws
  11. Discrete Law Calculations and Model Limits
  12. Normal Distribution and Standardization
  13. Normal Approximation and Interpretation
  14. Binomial and Normal Tail Approximation
  15. Estimating a Population Proportion
  16. Testing a Population Proportion
  17. Malnutrition Thresholds and Sample Size

1. Solving Statistical Exercises

★ Must-know

  • 🔄 A statistical exercise is handled in five steps:
    1. Read the data and identify the quantity sought.
    2. Choose the relevant concept, distribution, or formula.
    3. Check the conditions for using it.
    4. Calculate while retaining intermediate results.
    5. Conclude clearly and check the order of magnitude.

📌 Descriptive variance divides by nn, whereas estimated sample variance divides by n−1n-1.

Further detail

  • A statistical conclusion should identify the population, the measured quantity, and the meaning of the result; a number alone does not answer the question.

  • In the normal-distribution notation N(μ;σ)N(\mu;\sigma) used here, the second parameter is the standard deviation σ\sigma, not the variance.

  • For the normal table, Φ(z)=P(Z≤z)\Phi(z)=P(Z\le z); round zz to two decimal places, read four decimal places, and use Φ(−z)=1−Φ(z)\Phi(-z)=1-\Phi(z).

  • The two-sided normal critical values are u0.05=1.96u_{0.05}=1.96 and u0.01=2.576u_{0.01}=2.576; for a one-sided test at risk α\alpha, the threshold is u2αu_{2\alpha}, which is 1.645 at 5%.

  • The chi-square quantile χ1−α2(ν)\chi^2_{1-\alpha}(\nu) has order 1−α1-\alpha and ν\nu degrees of freedom; χ0.952(1)=3.841\chi^2_{0.95}(1)=3.841.

Memory Hook

Read, identify, check, calculate, conclude.

2. Descriptive Statistics and Survey Data

★ Must-know

📌 Qualitative nominal characteristics have unordered categories, qualitative ordinal characteristics have ordered categories, quantitative discrete characteristics are counts, and quantitative continuous characteristics are measurements.

  • A numeric code assigned to categories does not turn a qualitative characteristic into a quantitative one.

📐 Formula - For a class with width ai=ei−ei−1a_i=e_i-e_{i-1} and count nin_i, histogram density is di=niaid_i=\frac{n_i}{a_i}, so rectangle area aidia_i d_i equals the count.

📌 For an ordinal characteristic, use the mode, median, and frequencies rather than the mean of numeric category codes, because the mean changes with arbitrary coding.

Further detail

📐 Formula - For category frequency fi=ninf_i=\frac{n_i}{n}, the cumulative frequency is Fi=f1+⋯+fiF_i=f_1+\cdots+f_i and the corresponding sector angle is αi=360∘fi\alpha_i=360^\circ f_i.

📌 For unequal-width classes, the modal class is the class with the greatest density, not necessarily the greatest count.

📐 Formula - For grouped data, a quantile of proportion pp in the class spanning [ei−1,ei[[e_{i-1},e_i[ is estimated by ei−1+aip−Fi−1Fi−Fi−1e_{i-1}+a_i\frac{p-F_{i-1}}{F_i-F_{i-1}}; use p=0.5p=0.5 for the median and p=0.75p=0.75 for the third quartile.

  • The survey covers 240 road-accident victims admitted to a university hospital emergency department during three months; the observed individual is one admitted victim, so the results cannot automatically be generalized to all accident victims in the city.

  • In the survey, 168 of 240 victims were motorcyclists, giving a sector angle of 252°; the correct angle is calculated as 360∘nin360^\circ\frac{n_i}{n}, not by treating the count as degrees.

  • The survey’s grouped arrival-delay classes have widths 15, 15, 30, 60, and 120 minutes and respective counts 18, 42, 78, 66, and 36; their densities are 1.2, 2.8, 2.6, 1.1, and 0.3 victims per minute.

  • The arrival-delay median is estimated at 53.1 minutes by interpolation in the class [30, 60[; the median is a value, not the median class.

  • An estimated 57.5% of surveyed victims arrived within 60 minutes, and 28.75%, or about 69 victims, arrived more than 90 minutes after the accident.

  • The third quartile of arrival delay is estimated at 98.2 minutes; reaching the goal of 75% arriving within 60 minutes would require 180 arrivals by that time, 42 more than the observed 138.

Memory Hook

Ordinal categories have an order, but their numeric codes do not measure distances.

3. Counting Models and Combinatorial Principles

★ Must-know

📌 An arrangement counts ordered selections, whereas a combination counts selections without regard to order; replacement is possible in the with-replacement models and excluded in the without-replacement models.

📐 Formula - The six counting models are arrangements with replacement npn^p, arrangements without replacement Anp=n!(n−p)!A_n^p=\frac{n!}{(n-p)!}, permutations of distinct objects n!n!, permutations with repetitions n!n1!⋯nk!\frac{n!}{n_1!\cdots n_k!}, combinations without replacement Cnp=n!p!(n−p)!C_n^p=\frac{n!}{p!(n-p)!}, and combinations with replacement Cn+p−1pC_{n+p-1}^p.

  • Three counting principles apply:
    • Multiply for successive choices.
    • Add counts for mutually exclusive cases.
    • For “at least one,” subtract the cases with none from the total.

Further detail

  • The factorial convention is n!=n(n−1)⋯2⋅1n!=n(n-1)\cdots2\cdot1 and 0!=10!=1.

📌 When a counting problem includes a constrained position, fill that position first and then count the remaining choices.

  • For codes of three digits, the number containing at least one 7 is 103−93=27110^3-9^3=271, by subtracting codes containing no 7 from all codes.

Memory Hook

Order distinguishes arrangements from combinations; replacement distinguishes their variants.

4. Six Health-Service Counting Scenarios

★ Must-know

📌 Les situations 1 à 4 distinguent les résultats par leur ordre, tandis que les situations 5 et 6 ne distinguent pas l’ordre des éléments cités.

  • Les modèles des six situations sont respectivement un code avec répétition puis sans répétition, un arrangement sans répétition, une permutation avec répétition, une permutation de 7 objets distincts, une combinaison avec remise et une combinaison sans remise complétée par une somme et un produit.

Further detail

  • The five-digit pharmacy safe code uses digits 0–9 and cannot begin with 0.

  • A village tour selects five of eight villages in order and must include village V1, where a patient receiving tuberculosis treatment lives.

  • The word DREPANOCYTOSE has 13 letters, including two identical E’s, and the exercise counts its anagrams formed from exactly those letters.

  • Seven nurses, including three senior nurses, are assigned one each to seven distinct posts, and two specified intensive-care posts must be staffed by senior nurses.

  • The screening report records only the counts of 12 adults classified as normal, elevated, or hypertensive, and triggers an alert if at least half are hypertensive.

  • A mobile team of two interchangeable caregivers must consist either of two of the five physicians or of one of the four midwives and one of the six nurses.

  • Les six situations correspondent respectivement à un code de 5 chiffres dont le premier est non nul, une tournée ordonnée de 5 villages parmi 8, les anagrammes de DREPANOCYTOSE, l’affectation de 7 infirmiers à 7 postes, la répartition de 12 adultes entre 3 catégories et une équipe de 2 soignants de composition limitée.

  • Pour classer un dénombrement, on demande si l’ordre compte, si tous les objets sont placés, si les objets sont distincts et si un élément peut revenir.

5. Contraintes de dénombrement

★ Must-know

📐 Formula - Le nombre de codes de 5 chiffres dont le premier est non nul est 9×104=90 0009\times10^4=90\,000 lorsque la répétition est permise.

📐 Formula - Le nombre de codes de 5 chiffres distincts dont le premier est non nul est 9×9×8×7×6=27 2169\times9\times8\times7\times6=27\,216.

  • Il faut au minimum 6 chiffres distincts pour obtenir au moins 100 000 codes dont le premier n’est pas nul, car 5 chiffres donnent 27 216 codes et 6 en donnent 136 080.

📐 Formula - Parmi les tournées ordonnées de 5 villages parmi 8, A85=6 720A_8^5=6\,720 tournées sont possibles sans contrainte et 6 720−A75=4 2006\,720- A_7^5=4\,200 passent par V1.

📐 Formula - DREPANOCYTOSE contient 13 lettres, avec E et O répétés deux fois chacune, donc son nombre d’anagrammes est 13!2!2!=1 556 755 200\frac{13!}{2!2!}=1\,556\,755\,200.

📐 Formula - Pour affecter les 7 infirmiers aux postes en réservant 2 postes distincts à des seniors, on compte A32×5!=720A_3^2\times5!=720 affectations sur 7!=5 0407!=5\,040, soit une proportion de 17\frac17.

📐 Formula - Le nombre de comptes rendus donnant les effectifs de 12 adultes dans 3 catégories est (1412)=91\binom{14}{12}=91.

📐 Formula - Les équipes autorisées sont soit deux médecins, soit une sage-femme et un infirmier ; leur nombre est (52)+4×6=34\binom52+4\times6=34.

Further detail

📌 Dans un code sans répétition, le zéro interdit en première position redevient disponible pour les positions suivantes, contrairement à un décompte qui l’exclut à toutes les positions.

  • Avec d chiffres distincts et un premier chiffre non nul, passer de d à d+1 positions multiplie le nombre de codes par 10−d10-d.

📐 Formula - La proportion des tournées de 5 villages parmi 8 qui visitent V1 est 4 2006 720=58\frac{4\,200}{6\,720}=\frac58.

📐 Formula - Les anagrammes où les deux O se suivent sont comptées en traitant OO comme un bloc, ce qui donne 12!2!=239 500 800\frac{12!}{2!}=239\,500\,800 arrangements.

📐 Formula - La proportion des anagrammes de DREPANOCYTOSE où les deux O se suivent est 213≈0,154\frac{2}{13}\approx0{,}154, soit environ 15 %.

📐 Formula - Le nombre de comptes rendus déclenchant l’alerte, c’est-à-dire avec au moins 6 adultes en hypertension, est (86)=28\binom{8}{6}=28.

📌 Les 91 comptes rendus ne sont pas équiprobables lorsque chaque adulte est classé au hasard : les 3^12 suites individuelles sont équiprobables, mais le compte rendu (12,0,0) correspond à une suite et (4,4,4) à 34 650 suites.

📌 Le calcul (152)=105\binom{15}{2}=105 est faux pour les équipes autorisées, car il compte aussi les paires interdites de deux sages-femmes, deux infirmiers, ou un médecin associé à une sage-femme ou à un infirmier.

6. Probabilités conditionnelles et Bayes

Key Concepts & Definitions

  • Probabilité conditionnelle: La probabilité conditionnelle de A sachant B est P(A∣B)=P(A∩B)P(B)P(A\mid B)=\frac{P(A\cap B)}{P(B)} lorsque P(B)>0P(B)>0.
  • Système complet: Un système complet d’événements est constitué d’événements incompatibles deux à deux dont la réunion est certaine et dont les probabilités totalisent 1.

★ Must-know

📐 Formula - Si B1,…,BkB_1,\ldots,B_k forment un système complet, alors P(A)=∑i=1kP(Bi)P(A∣Bi)P(A)=\sum_{i=1}^{k}P(B_i)P(A\mid B_i).

📐 Formula - La formule de Bayes est P(Bj∣A)=P(Bj)P(A∣Bj)P(A)P(B_j\mid A)=\frac{P(B_j)P(A\mid B_j)}{P(A)}, où P(A)P(A) peut être calculée par les probabilités totales.

Further detail

  • Dans un arbre pondéré, on multiplie les probabilités le long de chaque branche et on additionne les probabilités des chemins menant à l’événement recherché.

Memory Hook

Les causes pondèrent les chemins vers l’effet ; Bayes remonte de l’effet vers la cause.

7. Mesures d’un test diagnostique

Key Concepts & Definitions

  • Sensibilité: La sensibilité est la probabilité d’un test positif parmi les malades, Se=P(T+∣M)Se=P(T+\mid M).
  • Spécificité: La spécificité est la probabilité d’un test négatif parmi les non-malades, Sp=P(T−∣¬M)Sp=P(T-\mid \neg M).
  • Valeur prédictive positive: La valeur prédictive positive est la probabilité qu’un individu soit malade sachant que le test est positif, VPP=P(M∣T+)VPP=P(M\mid T+).
  • Prévalence: La prévalence est la proportion de malades dans la population, notée p=P(M)p=P(M).
  • Valeur prédictive négative: La valeur prédictive négative est la probabilité qu’un individu ne soit pas malade sachant que le test est négatif, VPN=P(¬M∣T−)VPN=P(\neg M\mid T-).

Essential Points

📐 Formula - La valeur prédictive positive vérifie VPP=Se pSe p+(1−Sp)(1−p)VPP=\frac{Se\,p}{Se\,p+(1-Sp)(1-p)} et dépend de la prévalence autant que des performances du test.

Memory Hook

La sensibilité part des malades ; la VPP part des résultats positifs.

8. Dépistage des boîtes falsifiées

★ Must-know

  • Les parts des circuits et leurs taux de falsification sont respectivement:
    • centrale : 60 % des boîtes, 0,5 % falsifiées
    • grossistes : 30 % des boîtes, 2 % falsifiées
    • marché informel : 10 % des boîtes, 25 % falsifiées

📐 Formula - La probabilité qu’une boîte soit falsifiée est 0,60×0,005+0,30×0,02+0,10×0,25=0,0340{,}60\times0{,}005+0{,}30\times0{,}02+0{,}10\times0{,}25=0{,}034, soit 34 boîtes sur 1 000.

📐 Formula - Parmi les boîtes falsifiées, les probabilités d’origine centrale, grossistes et marché informel sont respectivement 3/34≈0,0883/34\approx0{,}088, 6/34≈0,1766/34\approx0{,}176 et 25/34≈0,73525/34\approx0{,}735.

  • Le test est positif pour 80 % des boîtes falsifiées et négatif pour 95 % des boîtes conformes, donc sa sensibilité est 0,80 et sa spécificité est 0,95.

  • Avec une prévalence de falsification de 0,034 et une sensibilité de 0,80, obtenir une VPP d’au moins 0,80 exige Sp≥0,9930Sp\geq0{,}9930, soit un taux de faux positifs d’au plus 0,704 %.

Further detail

  • Sur 10 000 boîtes, le test produit 272 vrais positifs, 68 faux négatifs, 483 faux positifs et 9 177 vrais négatifs.

📐 Formula - La probabilité qu’une boîte provienne du marché informel sachant que le test est positif est P(I∣T+)=0,10×0,23750,0755≈0,315P(I\mid T+)=\frac{0{,}10\times0{,}2375}{0{,}0755}\approx0{,}315.

📐 Formula - Après la fermeture du marché informel, les parts deviennent 2/32/3 pour la centrale et 1/31/3 pour les grossistes, la prévalence de falsification devient 0,010 et la VPP tombe à environ 0,139.

📌 La fermeture du marché informel réduit la prévalence de 3,4 % à 1 %, mais fait baisser la VPP de 0,360 à 0,139 sans modifier la sensibilité ni la spécificité.

9. Diagnostic Test Sensitivity and Specificity

★ Must-know

  • When prevalence is low, increasing specificity can improve positive predictive value more than increasing sensitivity: raising specificity from 0.95 to 0.99 raises it from 0.360 to 0.738, while raising sensitivity from 0.80 to 0.90 raises it to 0.388.

Further detail

  • With prevalence 0.034, sensitivity 0.90, and specificity 0.95, the positive predictive value is approximately 0.388, compared with 0.360 at sensitivity 0.80.

  • Among 10,000 wholesaler boxes with a falsification prevalence of 0.02, 200 are falsified, yielding 160 true positives and 490 false positives, so the positive predictive value is 160/650, approximately 0.246.

Memory Hook

Sensitivity raises detection; specificity restrains false positives.

10. Recognizing Discrete Probability Laws

Key Concepts & Definitions

  • Binomial distribution: Counts successes in a fixed number of independent trials with the same success probability; its parameters are n and p, and its possible values are 0 through n.
  • Poisson distribution: Counts events occurring individually and independently at a constant rate over a duration, with parameter λ and possible values 0, 1, 2, ….
  • Geometric distribution: Gives the rank of the first success in independent trials with the same success probability p, with possible values 1, 2, 3, ….
  • Hypergeometric distribution: Counts successes in n draws without replacement from a finite population of N individuals containing K successes, where p=K/N, and its possible integer values range from max(0, n−(N−K)) to min(n, K).

Essential Points

📐 Formula - For X∼B(n;p)X\sim B(n;p), P(X=k)=(nk)pk(1−p)n−kP(X=k)=\binom{n}{k}p^k(1-p)^{n-k}, E(X)=npE(X)=np, and V(X)=np(1−p)V(X)=np(1-p).

📐 Formula - For X∼P(λ)X\sim P(\lambda), P(X=k)=e−λλkk!P(X=k)=e^{-\lambda}\frac{\lambda^k}{k!} and E(X)=V(X)=λE(X)=V(X)=\lambda.

📐 Formula - For X∼G(p)X\sim G(p), P(X=k)=(1−p)k−1pP(X=k)=(1-p)^{k-1}p, E(X)=1/pE(X)=1/p, and V(X)=(1−p)/p2V(X)=(1-p)/p^2.

📐 Formula - For X∼H(N;n;p)X\sim H(N;n;p), P(X=k)=(Kk)(N−Kn−k)(Nn)P(X=k)=\frac{\binom{K}{k}\binom{N-K}{n-k}}{\binom{N}{n}} and E(X)=npE(X)=np, V(X)=np(1−p)N−nN−1V(X)=np(1-p)\frac{N-n}{N-1}.

📌 A binomial distribution can be approximated by a Poisson distribution with parameter np when n≥30 and p≤0.1.

📌 A hypergeometric distribution can be approximated by a binomial distribution when the sampling fraction n/N≤0.1.

  • To identify a discrete law, check in order whether the variable is the rank of the first success, counts events over a duration, or counts draws without replacement from a finite population; otherwise a fixed count of independent trials with constant probability is binomial.

Memory Hook

A fixed count suggests binomial; sampling without replacement suggests hypergeometric.

11. Discrete Law Calculations and Model Limits

★ Must-know

  • The example variances are below their means for binomial and hypergeometric counts, equal to their means for Poisson counts, and much greater than their means for geometric counts.

Further detail

  • For the fever-count example, 12 children are observed independently with fever probability 0.3, so the count follows B(12;0.3) and P(X≥3)≈0.7472.

  • For births averaging 3.2 per night under a constant independent event rate, the count follows P(3.2) and P(X≥5)≈0.2194.

  • For independent people with seropositivity probability 0.03, the rank of the first positive person follows G(0.03), and P(X>20)=0.97^{20}≈0.5438.

  • Drawing 6 students without replacement from a class of 30 containing 9 AS carriers gives a hypergeometric count, with P(X=2)=\binom{9}{2}\binom{21}{4}/\binom{30}{6}≈0.3629.

  • For 50,000 inhabitants each independently at risk 1/20,000, the exact weekly meningitis count follows B(50,000;0.00005), approximated by P(2.5), and P(X≤1)≈0.2873.

  • In a sample of 6 drawn without replacement from 30 students including 9 carriers, n/N=0.2>0.1, so a binomial model gives P(X=2)≈0.3241 instead of the exact hypergeometric value 0.3629.

  • For the weekly meningitis count, the smallest alert threshold with Poisson-model upper-tail probability at most 5% is 6 cases: P(X≥5)≈0.1088, while P(X≥6)≈0.0420.

  • In the meningitis example, contagion and epidemics can violate independence and cause a Poisson model to underestimate large case counts.

12. Normal Distribution and Standardization

Key Concepts & Definitions

  • Normal distribution: A continuous bell-shaped distribution symmetric about its mean μ; its parameter σ is the standard deviation, and P(X=a)=0 for any single value a.

★ Must-know

📐 Formula - If X∼N(μ;σ)X\sim N(\mu;\sigma), then Z=X−μσ∼N(0;1)Z=\frac{X-\mu}{\sigma}\sim N(0;1) and P(X≤x)=Φ(x−μσ)P(X\le x)=\Phi\left(\frac{x-\mu}{\sigma}\right), where Φ(z)=P(Z≤z)\Phi(z)=P(Z\le z).

Further detail

📐 Formula - For the standard normal distribution, Φ(−z)=1−Φ(z)\Phi(-z)=1-\Phi(z) and P(a<X<b)=Φ(b−μσ)−Φ(a−μσ)P(a<X<b)=\Phi\left(\frac{b-\mu}{\sigma}\right)-\Phi\left(\frac{a-\mu}{\sigma}\right).

  • The central 95% interval of a normal distribution is μ±1.96σ, corresponding to 2.5% in each tail.

  • The tabulated two-tailed quantiles are uα=1.282, 1.645, 1.960, 2.326, and 2.576 for α=0.20, 0.10, 0.05, 0.02, and 0.01, respectively.

  • For a normal distribution with μ=390 L/min and σ=50 L/min, the central 95% interval is [292,488] L/min.

Memory Hook

Picture a bell centered at μ, with z marking how many σ steps a value lies from its center.

13. Normal Approximation and Interpretation

★ Must-know

📌 A binomial variable with np≥5 and n(1-p)≥5 can be approximated by a normal variable with mean np and standard deviation √(np(1-p)).

  • When p is small, the binomial distribution can be right-skewed with a heavier right tail than its normal approximation, so the normal approximation may understate high-count probabilities.

Further detail

  • For 400 students with probability 0.0359 of DEP below 300 L/min, N follows B(400;0.0359), has mean 14.36, and its normal approximation is N(14.36;3.72).

  • Using continuity correction, P(N≥22)≈P(Y≥21.5)≈0.0274, compared with the exact binomial probability 0.0336.

📌 An observation with a small probability under a model is unusual but not impossible, and by itself does not prove that the model is false.

Memory Hook

A small p makes the binomial right tail heavier than the normal curve, so the normal approximation can understate rare high counts.

14. Binomial and Normal Tail Approximation

Essential Points

  • For the binomial distribution B(400; 0.0359) and the normal curve N(14.36; 3.72), the binomial bars for counts of 22 or more exceed the normal curve.

Memory Hook

In the right tail, the binomial bars rise above the normal curve.

15. Estimating a Population Proportion

Key Concepts & Definitions

  • Point Estimate: The unknown population proportion p is estimated by the observed sample frequency p^=xn\hat p = \frac{x}{n}, where x of the n sampled subjects have the characteristic.

★ Must-know

📐 Formula - The standard error for a confidence interval is p^(1−p^)n\sqrt{\frac{\hat p(1-\hat p)}{n}}, and the interval at confidence level 1−α1-\alpha is p^±uαp^(1−p^)n\hat p \pm u_\alpha\sqrt{\frac{\hat p(1-\hat p)}{n}}.

📌 To obtain a 95% confidence interval with half-width at most e, choose n≥uα2p(1−p)e2n \geq \frac{u_\alpha^2p(1-p)}{e^2}; without prior information about p, use p=0.5p=0.5 because p(1−p)p(1-p) is then maximal.

Further detail

📌 A normal-approximation confidence interval for p requires n≥30n \geq 30, np^≥5n\hat p \geq 5, and n(1−p^)≥5n(1-\hat p) \geq 5.

  • For a 95% confidence interval, u0.05=1.96u_{0.05}=1.96, and for a 99% confidence interval, u0.01=2.576u_{0.01}=2.576.

  • For a 95% confidence interval with half-width 0.05 and assumed proportion 0.5, the required sample size is at least 385 subjects.

Memory Hook

For an interval, use the observed proportion; for a test, use the null value.

16. Testing a Population Proportion

★ Must-know

  • A test of a proportion proceeds by stating H0 and H1, choosing the risk and checking conditions, calculating the test statistic, defining the critical region, making a decision and reporting the significance level, then concluding in context.

📌 Writing u_α=Φ⁻¹(1−α), for a one-sided upper-tail test, the critical region is z>u_α and the p-value is 1−Φ(z); for a one-sided lower-tail test, the critical region is z<−u_α and the p-value is Φ(z). For a two-sided test, the critical region is |z|>u_{α/2} and the p-value is 2[1−Φ(|z|)]. The significance level is the prespecified α.

📌 Failing to reject H0 does not prove H0; it means only that the data are insufficient to decide against it.

Further detail

  • At risks of 5% and 1%, the one-sided thresholds are 1.645 and 2.326, while the two-sided thresholds are 1.960 and 2.576.

📌 The direction of a one-sided test must be fixed before collecting data, because choosing it after seeing the results doubles the actual risk of a false alarm.

Memory Hook

Hypotheses → risk and conditions → statistic → critical region → decision → conclusion.

17. Malnutrition Thresholds and Sample Size

★ Must-know

  • The alert threshold for acute malnutrition prevalence is 10%, and the emergency threshold is 15%.

Further detail

  • In a sample of 480 children aged 6–59 months, 62 had acute malnutrition, giving p^=62/480≈0.12917\hat p=62/480\approx0.12917, or 12.9%.

  • For the sample of 480 children, the 95% confidence interval for prevalence is [9.9%; 15.9%], and the 99% interval is [9.0%; 16.9%].

  • Testing whether prevalence exceeds the 10% alert threshold gives z≈2.13z\approx2.13 and a one-sided significance level of 0.0166; the null is rejected at 5% but not at 1%.

  • Testing whether prevalence is below the 15% emergency threshold gives z≈−1.28z\approx-1.28 and significance level 0.1003, so the data do not establish that prevalence is below 15%.

  • If the observed proportion remains 62/480, a one-sided test against the 10% threshold rejects at the 1% risk when the sample size is at least 573.

  • For a 95% confidence interval with half-width at most 2.5 percentage points, using the observed proportion 62/480 requires at least 692 subjects; using p=0.5 without prior information requires 1,537 subjects.

📐 Formula - For a fixed observed difference from the null proportion, the test statistic grows in proportion to the square root of the sample size, z∝nz\propto\sqrt n.

Memory Hook

A larger sample makes the same observed difference more significant because z grows with √n.

Synthesis Tables

Mesures du test diagnostique

MesureConditionnementInterprétation
SensibilitéP(T+ | malade)Part des malades testés positifs
SpécificitéP(T− | non-malade)Part des non-malades testés négatifs
VPPP(malade | T+)Part des positifs réellement malades
VPNP(non-malade | T−)Part des négatifs réellement non-malades

Choosing a Discrete Distribution

DistributionWhat it countsKey condition
BinomialSuccesses in n trialsIndependent trials; constant p
PoissonEvents over a durationIndependent events; constant rate
GeometricRank of first successIndependent trials; constant p
HypergeometricSuccesses in finite-population drawsSampling without replacement

Test yourself

Take a 11-question practice quiz on Combinatorics and Diagnostic Testing, with every answer explained.

1. A student is analyzing a dataset and must report a result. Which statements describe appropriate choices in solving the exercise?

2. A researcher consults the normal table for a negative standardized value and selects test thresholds. Which statements are correct?

Take the quiz →

Study with flashcards

Memorize the key concepts of Combinatorics and Diagnostic Testing with 10 flashcards.

What five steps are used to handle a statistical exercise?

Read the data and identify the quantity sought, choose a course tool, check its conditions, calculate while retaining intermediate results, and conclude clearly while checking plausibility.

What distinguishes nominal, ordinal, discrete, and continuous characteristics?

Nominal categories are unordered, ordinal categories are ordered, discrete values are counts, and continuous values are measurements.

Does assigning numeric codes make qualitative categories quantitative?

No, numeric codes do not change a qualitative characteristic into a quantitative one.

See the flashcards →

Related classes

Make your own study guides

Upload your class notes and Revizly writes a study guide, a quiz and flashcards in about a minute.

Study guide maker