If an estimator $\theta$ is inconsistent, can I always conclude that $\theta$ is also biased?
$\begingroup$
$\endgroup$
-
4$\begingroup$ You can find a counterexample on Wikipedia: en.wikipedia.org/wiki/… $\endgroup$ – Tim♦ Sep 29 '17 at 13:29
$\begingroup$
$\endgroup$
To explain the relation between the bias and inconsistence, let's take a look at their mathematical definitions:
The definition of the bias is:
$Bias(\theta)=E(\hat{\theta})-\theta$
whis is the expected value of the estimator $\hat{\theta}$ minus the true parameter $\theta$.
Furthermore, one calls an estimator inconsistent in mean squared error, if
$lim_{ (N \to \infty )}MSE(\hat{\theta})=Bias(\hat{\theta})^2+Var(\hat{\theta}) \neq 0$
holds. ($N$ denotes the sample size)
Recall, that often the convergence in probability is used to check consistency, which is given by:
$plim_{N \to \infty} \hat{\theta}=\theta$
Both versions refer to the asymptotic behaviour of $\hat{\theta}$ and expresses that, as data accumulates, $\hat{\theta}$ gets closer and closer to the true value of $\theta$. This argumetation is outligned in: http://www.stats.ox.ac.uk/~steffen/teaching/bs2siMT04/si2c.pdf
Now to answer the question "Is an estimator allways biased if he is inconsistent?" just look at the formula: If the estimator is inconsistent, this might be due to a zero Bias(=unbiased) but a non-zero variance. Therefore, an estimator might be inconsistent but unbiased.
-
3$\begingroup$ Your definition of inconsistency looks unusual. Ordinarily the definition concerns convergence in probability. Please see stats.stackexchange.com/questions/31036/…, especially the long comment thread following one answer. $\endgroup$ – whuber♦ Sep 29 '17 at 18:37
-
1$\begingroup$ @whuber thanks for the comment. I edited the aswer and included two approaches on how to check on consistency. $\endgroup$ – Jogi Sep 29 '17 at 20:17
