FEATURED ARTICLE

Introduction to Statistical Inferences Making

BIS.Net Team

Introduction

Statistical Inference is required in every aspect of life by every creature. There are many complex definitions of the meaning of inferences. Many of these are philosophical and academic making it hard for the every day person to understand the relevance and importance to decision making.

A good way to understand what a statistical inference is through example. We make inferences from observations. Another way of expressing this is by saying we draw conclusions based on our observations. But observational statistical inference making is prone to error. By observing that birds have wings and that they fly we some will incorrectly conclude that life with wings can fly and life without wings cannot fly. When a pig is seen, based on observations some will conclude that pigs cannot fly.

Statistical inferences can be wrong

Statistical Inferences and conclusions can be wrong. Not all life with wings can fly. Some birds, such as Emus and Ostriches, cannot fly even though they have wings. Some life without wings can fly. Man can fly without wings. Although a plane has wings, a helicopter has no wings (in the normal sense). A further complication is the definition of wings, e.g. feathered or non-feathered.

The lesson is that inferences can be wrong. There is always the risk that our observations are not exhaustive enough.

BIS.Net Analyst is about making inferences based on observed data. Such inferences making is applicable in many decision-making situations. Politicians infer voter sentiments based on surveys. Doctors infer the patient’s risk of heart attach from blood pressure measurements. The QA manager makes a statistical inference on non-conformances based on samples taken from a lot.

All inferences and hence conclusions and thus decisions can be wrong. There are many reasons for these errors. Insufficient observations, jumping to conclusions, non-random sampling, insufficient sampling are a few examples.

Random sampling error is a major source of error. If we sample 10 cans of Coca-Cola from a batch of 10000 cans and obtain the average net volume from the sample that net volume will most likely not be equal to the net volume of all the cans in the batch. Another sample would most likely provide a different estimate

Yet, even today, sampling error is rarely considered in a professional manner. An operator (not all) will take a sample of five and decide whether the process is on target by assuming the sample average is equal to the true average. A doctor (not all) will prescribe blood pressure medication based on three measurements.

Tools for more reliable statistical inferences

BIS.Net Analyst provides tools to enable professionals to make reliable statistical inferences. Two such important tools are confidence intervals and hypothesis tests.

Even though, the scientific approach is also not perfect. Because inferences are based on samples, there is always a risk of drawing false conclusions. However these risks are controlled by specifying the risk of falsely concluding there is an effect when there is none, or false concluding there is no effect, when there is one.

Additionally BIS.Net Analyst uses machine-powered algorithms in many instances, to not only improve reliability, but make statistical inferences possible where previously it was not with classical approaches, such as hypothesis testing for capability indexes.