Problem specification

There is a product and a measurement process:

  • Product ~ $N(\mu_P, \sigma_P)$
  • Measurement ~ $N(\mu_M, \sigma_M)$ (the measurement error is independent of the product under investigation)

I wish to characterize the product, i.e. estimate its mean. To this end, $n_P$ products are measured $n_M$ times. Please note that each of the $1, ..., n_P$ products is measured $n_M$ times. Therefore the measurements are not independent but correlated.


Regarding the SEM, I wish to obtain an expression of the form $$\sigma_{\bar{z}}=\sigma_{\bar{z}}(n_P, n_M; \sigma_P, \sigma_M)=\textrm{your input}$$ to then make a qualified decision on how to properly chose $n_P$ and $n_M$ to satisfy certain accuracy requirements.

Proposed solution

Let $Z$ denote the random variable of the mean, $X$ the random variable of product and $Y$ the random variable of the measurement. This gives $$ Z(n_P, n_M) = \frac{1}{n_P n_M}\sum_{i=1}^{n_P}\sum_{j=1}^{n_M} X_i + Y_{i, j}. $$ Comuting the variance of $Z$ gives $$ Var(Z) = \frac{1}{(n_P n_M)^2}\sum_{i=1}^{n_P} Var(n_M X_i) + \sum_{j=1}^{n_M} Var(Y_{i, j}) \\ = \frac{1}{(n_P n_M)^2}\sum_{i=1}^{n_P} n_M^2 Var(X_i) + n_m Var(Y_{i, j}) $$ and after simplifying, since $Var(X_i) = Var(X) = \sigma_P^2$ and $Var(Y_{i,j}) = Var(Y) = \sigma_M^2$, we obtain $$ \sigma_{\bar{z}}^2 = Var(Z) = \frac{1}{n_P}\left(\sigma_P^2 + \frac{\sigma_M^2}{n_M}\right). $$

Concerning the confidence interval: $$\bar{z} \pm t_{df;1-\alpha/2} \times s_{\bar{z}}$$ with $df = n_P (1+n_M) -1$.

Are these approaches correct?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.