Further analysis of the deseasonalised data
The first step in analysing seasonal data is usually to remove the seasonal effect.
Data = Seasonal effect + Deseasonalised
However we should next examine the deseasonalised data for trend and cyclical features. Using the methods in the first two sections of this chapter, we can obtain trend and cyclical components. What is left after removing these are the residuals.
Deseasonalised = Trend + Cyclical + Residual
Putting these together, we can identify four separate components to the variation in our original time series.
Data = Seasonal effect + Trend + Cyclical + Residual
It is important that you understand the distinction between these four components of a seasonal time series:
Monthly temperatures in Boulder, Colorado
You will use the diagram below to help understand the different components of the time series of mean monthly temperature in Boulder Colorado.
Initially, the blue line in the top half of the diagram connects the raw data points. Click the checkbox Seasonal to delete the seasonal effect from this diagram and display it in the bottom half of the diagram. (The mean has been added to the seasonal effects to allow them to be plotted against the same axis as the data.) The top diagram therefore shows the deseasonalised data and the bottom shows the seasonal effects.
There is virtually no trend or autocorrelation in the deseasonalised data. However also click the checkboxes Trend and Cyclical to remove these components from the top series and add them to the bottom. (There is very little change.) What remains in the top is the residuals.
Selecting the components one-at-a-time allows them to be displayed separately in the bottom half of the diagram.