Let $\{\hat{X}_n\}$ be a sequence of estimators that converges in probability to the constant $\bar{x}$, which I take to mean that, for any $\epsilon > 0$, $\lim \limits_{n \to \infty} \Pr(|\hat{X}_n - \bar{x}| > \epsilon) = 0$.

I have also seen people write $(\hat{X}_n - \bar{x}) = o_p(1)$ and have seen some people refer to it as equivalent to the statement that $\hat{X}_n$ is consistent for $\bar{x}$.

What is the difference between these two statements?

Is it correct that $(\hat{X}_n - \bar{x}) = o_p(1)$ is equivalent to $\lim \limits_{n \to \infty} \Pr(|\hat{X}_n - \bar{x}| \geq \epsilon) = 0$? And could one also write that $(\hat{X}_n - \bar{x}) = o_p(1)$ is equivalent to $\hat{X}_n = \bar{x} + o_p(1)$?

Thank you for any and all help!


2 Answers 2


Yes, $X_n-\bar x=o_p(1)$ is just notation for saying "For every $\epsilon>0$, $$\lim_{N\to\infty}P\left(|X_n-\bar x|>\epsilon \right)=0"$$

More generally, $X_n-Z_n=o_p(Y_n)$ is notation for saying "For every $\epsilon>0$, $$\lim_{N\to\infty}P\left(|X_n-Z_n|>\epsilon |Y_n| \right)=0"$$


And could one also write that $(\hat{X}_n - \bar{x}) = o_p(1)$ is equivalent to $\hat{X}_n = \bar{x} + o_p(1)$?

I do not know whether this is conventional, but you can do it.

Interpretation in terms of quantile functions

We can argue about these notations with $o_p$ and $O_p$ in the same way as for $o$ and $O$ when we make an interpretation that connects the two.

With a particular interpretation you can turn the probabilistic expressions ${o}_p(f(n))$ or ${O}_p(f(n))$ into an expression with the more well known ${o}(f_n)$ or ${O}(f_n)$.

We can interpret

$$X_n = {o}_p(f(n)) \quad \text{for }n \to \infty$$

in terms of the quantile functions for $X_n$

$$\forall p: \lbrace Q_{X_n}(p) = {o}(f(n)) \quad \text{for }n \to \infty \rbrace$$

which means

$$\forall p: \lim_{n \to \infty} \frac{Q_{X_n}(p)}{f(n)} = 0$$

For example: in the figure below we show some quantiles functions of the mean $\bar{X}_n = \frac{1}{n}\sum_{i=1}^n X_i$, where $X_i \sim N(\mu,\sigma^2)$ with $\mu = 0.5$ and $\sigma =1$

convergence of quantile functions

The convergence of the quantile functions means that for every for any $0 < p \leq 1$, there is an $n$ such that $Pr(\vert X_n \vert > \epsilon) \leq p $

Interpretation of equal sign

With this view, in terms of quantiles, the notations $o_p$ and $O_p$ follow the same rules as the notations $o$ and $O$ in 'regular' mathematics.

For these notations, there are some debates on how the equality sign '$=$' must/can be interpreted. See for instance

The equality sign '$=$' is not the typical equality and should be interpreted as a short-handed notation for 'is element of', or in terms of symbols '$\in$', and should not be interpreted as an 'identity'.

The acceptance and use of notations with an algebraic appearance like $\hat{X}_n = \mu_{x} + o_p(1)$ is a matter of taste, but it is certainly not wrong. You may however need to keep in mind that you do not confuse the reader.

To contemplate the meaning of $\hat{X}_n = \mu_{x} + o_p(1)$, the reader might need to imagine eventually the 'difference' $\hat{X}_n - \mu_{x} = o_p(1)$. So you could write it out like that from the start.

Which notation you use might depend on the intuition that you want to stress.

  • The function is some other function plus/minus some bounds: $$f(x) = g(x) + O(h(x))$$
  • The difference between the function and some other function is within some bounds: $$f(x) - g(x) = O(h(x))$$

To me it is a nuanced difference and the notations like $f(x) = g(x) + O(h(x))$ are not wrong. They are actually being used a lot (this sounds like an argumentum ad populum, but it is not a fallacy when we are arguing about conventions and use of interpretations that might be confusing when they are uncommon).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.