Copyright © Vassilis Hajivassiliou, LSE 1998-2021
Experiments on Central Limit Theorems
The results of five sets of experiments are overviewed here, characterized
by five true underlying distributions -- populations.
After reviewing these, you will be given the chance to
try experimenting interactively.
- Experiment 1: samples drawn from a normal distribution.
Empirical (Monte-Carlo) p.d.f.'s
- Experiment 2: we draw samples from an exponential distribution which is
positively skewed.
We observe that the CLT still works for large sample
sizes and produces an approximately bell-shaped distribution for the
standardized mean, despite the underlying skewness.
Empirical (Monte-Carlo) p.d.f.'s
- Experiment 3: we work
with a binomial distribution and the sample proportion of successes.
The CLT case then illustrates the De Moivre --- Laplace theorem.
Observe that the ``gaps" due to the underlying discreteness are filled as
the sample size grows --- in the limit the distribution becomes normal.
Empirical (Monte-Carlo) p.d.f.'s
Missing File
- The
last two sets of experiments highlight the importance of the necessary
assumptions for the LLN's and CLT's to work.
- Experiment 4:
Here the observations are drawn from a Cauchy distribution,
which has no finite moments. As a result, both the LLN and the CLT fail.
Empirical (Monte-Carlo) p.d.f.'s
- Experiment 5:Here the Pareto distribution is used to generate the draws. A
single parameter of this distribution determines whether the variance will
be infinite (in which case the CLT fails to hold) and/or the mean will be
infinite (leading to a failure of the LLN as well).
-
THETA=3 --- Finite Mean, Finite Variance.
Empirical (Monte-Carlo) p.d.f.'s
-
THETA=1.5 --- Finite Mean, Infinite Variance.
Empirical (Monte-Carlo) p.d.f.'s
-
THETA=0.5 --- Infinite Mean (& Infinite Variance).
Empirical (Monte-Carlo) p.d.f.'s
Run your own CLT Experiments Interactively.
Back to Overview