Find a simple discrete model for X with E[X]=1.2 and Var(X)=2.5.

You're on a roll — The problem asks to find a simple discrete model for the number of defective items per shipment, X, given E[X] = 1.2 and Var(X) = 2.5. Step 1: Analyze the properties of common discrete distributions. • Poisson distribution: For a Poisson distribution with parameter , E[X] = and Var(X) = . Given E[X] = 1.2 and Var(X) = 2.5. If we use a Poisson distribution, we would need = 1.2 and = 2.5 simultaneously, which is not possible. Also, the variance (2.5) is not equal to the mean (1.2). So, a Poisson distribution is not a good fit. • Binomial distribution: For a binomial distribution with parameters n and p, E[X] = np and Var(X) = np(1-p). Given E[X] = 1.2 and Var(X) = 2.5. From E[X] = np = 1.2. From Var(X) = np(1-p) = 2.5. Substitute np = 1.2 into the variance equation: 1.2(1-p) = 2.5. 1 - p = (2.5)/(1.2) ≈ 2.083. p = 1 - 2.083 = -1.083. Since p must be between 0 and 1, a binomial distribution is not a good fit. • Bernoulli distribution: A Bernoulli distribution is a special case of a binomial distribution where n=1. For a Bernoulli distribution with parameter p, E[X] = p and Var(X) = p(1-p). Given E[X] = 1.2. This would imply p = 1.2, which is not possible as p must be between 0 and 1. So, a Bernoulli distribution is not a good fit. Step 2: Consider a custom two-point distribution. A two-point distribution assigns probabilities to two distinct values. Let these values be x_1 and x_2 with probabilities p_1 and p_2 respectively, where p_1 + p_2 = 1. The mean is E[X] = x_1 p_1 + x_2 p_2. The variance is Var(X) = E[X^2] - (E[X])^2 = (x_1^2 p_1 + x_2^2 p_2) - (x_1 p_1 + x_2 p_2)^2. Let's check option D: "Use a custom two-point distribution where X=1 or X=2 with chosen probabilities that match the sample mean and variance." If X can only take values 1 or 2, let P(X=1) = p and P(X=2) = 1-p. Then E[X] = 1 · p + 2 · (1-p) = p + 2 - 2p = 2 - p. We are given E[X] = 1.2. So, 2 - p = 1.2 p = 2 - 1.2 = 0.8. Thus, P(X=1) = 0.8 and P(X=2) = 1 - 0.8 = 0.2. Now, let's calculate the variance for this distribution: E[X^2] = 1^2 · p + 2^2 · (1-p) = 1 · 0.8 + 4 · 0.2 = 0.8 + 0.8 = 1.6. Var(X) = E[X^2] - (E[X])^2 = 1.6 - (1.2)^2 = 1.6 - 1.44 = 0.16. This calculated variance (0.16) does not match the given Var(X) = 2.5. Therefore, a two-point distribution with X=1 or X=2 is not the correct model. Step 3: Re-evaluate the options based on the problem statement. The question asks for a "simple discrete model consistent with these estimates for internal communication (even if not uniquely correct)". This implies we are looking for the best fit among the given options, even if no option perfectly matches all properties. Let's re-examine the options: A. Use a Poisson distribution with parameter = 1.2, since Poisson has E[X] = Var(X) = , matching the approximate mean and variance. This statement is incorrect because E[X] = 1.2 and Var(X) = 2.5. They are not equal, so a Poisson distribution is not a good fit. B. Use a binomial distribution with parameters n=1 and p=1.2 so the mean matches the observed 1.2 defects per shipment. This is a Bernoulli distribution (n=1). As shown in Step 1, p=1.2 is not a valid probability. So, this option is incorrect. C. Use a Bernoulli distribution with parameter p=1.2 because the mean of a Bernoulli is p. This is the same issue as option B; p=1.2 is not a valid probability. So, this option is incorrect. D. Use a custom two-point distribution where X=1 or X=2 with chosen probabilities that match the sample mean and variance. As shown in Step 2, if we choose X=1 or X=2, we can match the mean (E[X]=1.2) but not the variance (Var(X)=0.16 ≠ 2.5). However, the option states "with chosen probabilities that match the sample mean and variance". This implies that such probabilities could exist for some two-point distribution, not necessarily restricted to X=1 or X=2. Let's consider a general two-point distribution with values x_1 and x_2 and probabilities p and 1-p. E[X] = x_1 p + x_2 (1-p) = 1.2 Var(X) = E[X^2] - (E[X])^2 = (x_1^2 p + x_2^2 (1-p)) - (1.2)^2 = 2.5 x_1^2 p + x_2^2 (1-p) = 2.5 + 1.44 = 3.94 We have two equations and four unknowns (x_1, x_2, p, 1-p). We need to choose x_1 and x_2 such that a valid p exists. Let's try to find x_1, x_2 and p that satisfy these conditions. From E[X] = x_1 p + x_2 (1-p) = 1.2. From Var(X) = (x_1 - E[X])^2 p + (x_2 - E[X])^2 (1-p) = 2.5. (x_1 - 1.2)^2 p + (x_2 - 1.2)^2 (1-p) = 2.5. Let's try to solve for p in terms of x_1 and x_2: p = (1.2 - x_2)/(x_1 - x_2). Substitute this into the variance equation. This will be a complex equation. However, the question asks for the best simple discrete model. The key observation is that Var(X) > E[X] (2.5 > 1.2). • For a Poisson distribution, Var(X) = E[X]. • For a Binomial distribution, Var(X) = np(1-p) = EX. Since 0 < p < 1, 0 < 1-p < 1, so Var(X) < E[X]. Therefore, neither Poisson nor Binomial distributions can model this data because they both have variance less than or equal to the mean. A distribution where the variance is greater than the mean is called overdispersed. A two-point distribution can be overdispersed. Let X take values x_1 and x_2 with probabilities p and 1-p. E[X] = x_1 p + x_2 (1-p) Var(X) = (x_1 - E[X])^2 p + (x_2 - E[X])^2 (1-p) This can be rewritten as Var(X) = p(1-p)(x_1 - x_2)^2. We need p(1-p)(x_1 - x_2)^2 = 2.5 and x_1 p + x_2 (1-p) = 1.2. We can choose x_1 and x_2 such that these conditions are met. For example, if we choose x_1=0 and x_2=k: E[X] = k(1-p) = 1.2 1-p = 1.2/k. Var(X) = p(1-p)(0-k)^2 = p(1-p)k^2 = 2.5. Substitute 1-p = 1.2/k: p(1.2/k)k^2 = 2.5 1.2pk = 2.5 p = (2.5)/(1.2k). We also have p + (1-p) = 1, so (2.5)/(1.2k) + (1.2)/(k) = 1. Multiply by 1.2k: 2.5 + 1.44 = 1.2k. 3.94 = 1.2k k = (3.94)/(1.2) ≈ 3.283. Then p = (2.5)/(1.2 × 3.283) ≈ (2.5)/(3.94) ≈ 0.6345. And 1-p = 1 - 0.6345 = 0.3655. Check: E[X] = 0 · p + 3.283 · (0.3655) ≈ 1.2. Check: Var(X) = p(1-p)(0-3.283)^2 = 0.6345 · 0.3655 · (3.283)^2 ≈ 0.2318 · 10.778 ≈ 2.5. So, a two-point distribution can match the given mean and variance. The specific values X=1 or X=2 mentioned in option D are just an example, and the general statement "with chosen probabilities that match the sample mean and variance" is the key. Therefore, a custom two-point distribution is the only type of distribution among the options that can accommodate Var(X) > E[X]. The final answer is D. Just one more to go! Send it over.