09. Decomposing data

Decomposing Data

Decompose - to break down into components.

(Merit)

Decomposing data separates it into the Trend, Seasonal, & Residual component.

The relative influence of each of the components (as a percentage of the overall variation) can be investigated.

What is the total variation of the Data? (over the time period)

• • What is the Trend Component (% of the total variation of the Data)

  • What is the Seasonal Component (% of the total variation of the Data)

  • What is the Residual Component (% of the total variation of the Data)

How do these compare?

Discussion MUST be in CONTEXT

'As a guide' a good model has the residual component of variation is less than 10% of the total variation.

Residuals - Are there any unusual observations - do they warrant further investigation.

Note: iNZight graphs do not have labels on the vertical axis. For Merit or better you will need to add a lable with units on the vertical axis.

Discuss relative effect of seasonal effect vs long term trend using numerical values

What is the Total Variation of the Data? (over the time period) 13 000 visitors

What is the Trend Component :

38% - Less of an influence over total variation than the Seasonal component

What is the Seasonal Component :

62% - Cause of most of the variation in the number of visitors from Germany (Remember the context)

What is the Residual Component :

8% - This is below 10% indicating that the iNZight Additive model is a reasonably good model for this situation

Residuals - Are there any unusual observations - do they warrant further investigation.

Calculations Overview (you tube video by Pricilla Allan)