[Home]

Table of contents


Exercises on identifiability

  1. Consider the model $y_{ij} = \mu + \alpha_i + \epsilon_{ij},$ where $i=1,2,3$ and $j=1,...,5.$ We assume the Gauss-Markov set up. State with reason which of the following parametric functions are estimable:
    1. $\alpha_1-\alpha_3$
    2. $\mu+\alpha_1$
    3. $\mu-\alpha_2$
    4. $\alpha_1-2 \alpha_2+\alpha_3$
  2. Same set up as above. Is it true that all parametric functons of the form $\sum_i c_i \alpha_i$ with $\sum c_i = 0$, are estimable. Also, is it true that all such parametric functons must be non-estimable if $\sum c_i\neq 0?$
  3. Consider the Gauss-Markovian linear model $y_{ijk} = \mu + \alpha_i + \beta_j + \epsilon_{ijk},$ where $i=1,...,I$ and $j=1,..,J$ and $k=1,...,n_{ij} (\geq 2).$ State with reason which of the following parametric functions are estimable:
    1. $\alpha_i-\alpha_j$ for $i\neq j.$
    2. $\alpha_i+\alpha_j$ for $i\neq j.$
    3. $\beta_i-\beta_j$ for $i\neq j.$
    4. $\mu+\alpha_i$
    5. $\mu-\alpha_i$
    6. $\mu+\beta_j$
    7. $\mu+\alpha_1+\beta_2$
  4. Same set up as in the last problem.
    1. If $\mu + \sum c_i \alpha_i$ is estimable:, then what can you say about $c_i$'s?
    2. If $\sum c_i \alpha_i$ is estimable:, then what can you say about $c_i$'s?
    3. If $\mu + \sum c_i \alpha_i + \sum d_j \beta_j$ is estimable:, then what can you say about $c_i$'s and $d_j$'s?
  5. Same set up as above. Find the dimension of the space of all identifiable linear parametric functions. How does the answer change if all the $n_{ij}$'s are known to be equal?

Problems from the book by Rao and Sengupta

In the following problems our familiar Gauss-Markov model is denoted by $(\vec y, X \vec \beta , \sigma^2 I).$
Rao Sengupta 1

Rao Sengupta 2

Rao Sengupta 3

Rao Sengupta 4

Rao Sengupta 5

Rao Sengupta 6

Rao Sengupta 7

Rao Sengupta 8

Rao Sengupta 9

The data set for the following problem is available in R under the name stackloss.
Rao Sengupta 10

Rao Sengupta 11