Product of Two Draws: With vs Without Replacement
From the numbers
Hints
- Express $E[XY]$ using the sum and sum of squares of $ through0$.
- With replacement the draws are independent: $E[XY] = \mu^2$. Without replacement they are negatively correlated.
- Without replacement, $E[XY] = (S^2 - Q)/(n(n-1))$, which is less than $\mu^2$ because it removes the squared (diagonal) terms.
Worked Solution
How to Think About It: Write the expected product in terms of the sum and the sum of squares. Let the draws be $X$ and $Y$. With replacement, $X$ and $Y$ are independent with the same mean $\mu$, so $E[XY] = \mu^2$. Without replacement, the two draws are negatively correlated (picking a high number first leaves a slightly lower-mean pool), which drags the expected product DOWN.
The Clean Argument: Let $S = 1 + \cdots + 10 = 55$ and $Q = 1^2 + \cdots + 10^2 = 385$, with $\mu = 5.5$.
- With replacement: $E[XY] = E[X]E[Y] = \mu^2 = 5.5^2 = 30.25$.
- Without replacement: by symmetry $E[XY] = \dfrac{(\sum_i i)^2 - \sum_i i^2}{10 \cdot 9} = \dfrac{S^2 - Q}{90} = \dfrac{55^2 - 385}{90} = \dfrac{3025 - 385}{90} = \dfrac{2640}{90} \approx 29.33.$
Why: With replacement is larger. Intuitively, drawing without replacement forbids the diagonal pairs $(i,i)$ -- in particular it removes the large product
0 \times 10 = 100$ and other big squares -- and introduces negative dependence, both of which lower the expected product.Answer: WITH replacement gives the larger expected product ($30.25$ versus $\approx 29.33$). The intuition: without-replacement draws are negatively correlated and exclude the high squared terms, so the product's expectation is smaller.
Intuition
The key conceptual point is the sign of the dependence: sampling without replacement induces negative correlation between draws, which lowers the expectation of their product relative to the independent case. You can see it without arithmetic -- with replacement allows the big square terms like
0^2$, while without replacement forbids drawing the same number twice, stripping out exactly those large diagonal contributions. This negative-dependence-from-finite-sampling effect matters in finite-population statistics, sampling-based estimators, and any setting where draws compete for a fixed pool.