![]() |
Time series
|
Use the 'Plot time series' button to generate 18 scatterplots. For each, the sample value (X - coordinate) is plotted against the sample value at a fixed time increment in the future (Y -coordinate):
In each plot the time increment is equal to the start lag plus the sum of the cumulative time steps (i.e., start lag + time step X plot number). The increment value is shown at the top of each plot. If there is no temporal predictability, the point distribution will be random. However, if temporal patterns exist, the distribution of points will be non-random, and the shape of the distribution will indicate the degree of symmetry and the scatter will indicate the degree of randomness or 'noise'. You can replot with different start lags and time step increments. Click the 'Print plots' button for hard-copy output. Note that 24, not 18, time steps are plotted, and that you cannot select this button until data have been plotted on the screen.
• Use the 'periodicity test… ' button to generate a summarized periodicity test for 50, 100, 200, or more stepped time intervals (NOTE: the interval selection buttons are only available if there are sufficient points within the block). For 200 or fewer steps, results are shown as a bar graph; for more steps a line plot is drawn. This is the setup screen for generating a periodicity test:
![]()
Output from a typical periodicity test looks like this:
![]()
This example shows a 750-step plot. Correlations with negative slopes are plotted below the zero line. Bar (or line) heights are a relative index of how value at time ( T ) predicts value at time ( T + time step). The tallest bar (red on color screens) is the interval with highest predictability. Click the 'Show Other Peaks Where...' button to display only those time correlations with r2 higher than a specified value. Click the 'Print' button for hard-copy output; click the 'Quit' button, another window, or the close box to return to the analysis page.
go to: |