Class 9: Perform & Communicate Diagnostic Analytics

What is the definition of diagnostic analytics?

Investigates the underlying cause that cannot be answered by simply looking at descriptive data

What are the Tools & Techniques of Diagnostic Analytics?

Hypothesis testing, variances, differences from expectations, correlations, visualize and cluster, identify outliers/anomalies, drill-downs and roll-ups, pivot tables, sequence check

What are the two broad types of Diagnostic Analytics?

1. Identifying Anomalies/Outliers

2. Finding Previously Unknown Linkages, Patterns, or Relationships between and Among Variables

What is often the firm step in diagnostic analytics?

To look for and identify unusual, unexpected results or transactions

What do you do with anomalies/outliers you identify?

• Investigate further to understand why they occurred

• Identify if it is an error or fraud, or just an extreme observation

How do you find patterns in the underlying data set?

• Summarize data at different levels and uncovering additional details to understand why something happened

• Drill Down Detailed Analysis

• Perform Statistical Analysis

What are examples of Diagnostic Analysis?

• Differences from Expectations

• Heat Maps

• Outlier Inspection

• Regression Analysis

What are accountants’ expectation on a specific issue?

Accountants expect certain levels of performance or outcomes, and when that doesn’t happen, it usually triggers an additional investigation to figure out what

What is Management by Exception?

A managerial style that allows management to spend its time addressing issues/problems

What is an example of a Diagnostic Tool?

Box Charts

What can Box Charts be used to do?

• Visualize the distribution of data points

• Identify outliers

What is the Whisker of a Box Chart?

The largest (upper) or lowest (lower) points within 1.5x the min or max of the IQR

What is the IQR of a Box Chart?

Difference between the upper and lower quartile

Does running multiple regressions fix the problem of outliers?

No

What can the same regression equation yield?

Opposite stories, depending upon the sample examined and the treatment of outliers

What are the two common ways of dealing with outliers?

• Trim the outlier

• Winsorize the outlier

What does it mean to Trim the Outlier?

• Removes the outlier completely

• Sample size decreases

What does it mean to Winsorize the Outlier?

• Pulls the outlier back into the distribution

• Retains the outlier observation, but changes its value

• Sample size remains the same, but the value of certain observation changes to be closer to the center of the distribution

What is the more sophisticated way to handle outliers?

Robust Regression

What is Benford’s Law?

A principle that predicts the frequency distribution of digits in naturally occurring datasets, where lower digits appear more frequently as the leading digit.

Where is Benford’s Law applied?

In analyzing the validity of statistics and financial records

Does applying Benford’s Law Analysis to the numbers of a firm’s financial statement prove whether managers manipulated their accounting info?

No

What are potential causes of outliers?

• natural variation

• data entry and measurement errors

• sampling problems

What is Drill-Down Testing in Diagnostic Analytics?

Drill into the analytics to help explain why or how

Does correlation mean causation?

No

What is a Type 1 Error?

False rejection of the null hypothesis

What is a Type 2 Error?

Falsely failing to reject the null hypothesis

What do we use to test a hypothesis?

Some type of statistical analysis, which allows us to have a framework for our tests

What does hypothesis testing gives us?

• Coefficients: tell us the effect size in economic significance

• T-Stats and P-Values: tell us the effect size in statistical significance

What is a P-Value?

The probability of observing a t-statistic as extreme as the one shown if the null hypothesis is true

What are small p-values indicative of?

That we can reject the null hypothesis of no relation

What arbitrary threshold do we often use when discussing statistical significance?

5%

What is a p-value really asking?

How rare would it be to see this data if the hypothesis were true?