home *** CD-ROM | disk | FTP | other *** search
- The data for the Quebec example come from Thwaites, J. (1984) "La grève
- au Québec: une analyse quantitative exploratoire portant sur la période
- 1896-1915." Labour/Le Travail 14: 183-204. The columns in the data file
- correspond to the Year, the number of Strikes, the number of Strikers, the
- Days Lost, the number of Offensive strikes, the number of Defensive strikes,
- and the total Length of the strikes.
-
- These are time series which Statpack can only handle as a regression
- problem, without taking into account any autocorrelations.
-
- A preliminary Plot of each variable against the Years shows that they
- all increase from the beginning to the middle of the period, then decrease
- into the war period.
-
- Although many models are possible, let us first consider the variables
- influencing Days Lost. In NReg, we shall start with each other variable in
- turn as independent, keeping those which are significant. Days Lost does not
- depend on the number of Strikes, but rather on the number of Strikers. We
- inspect the residuals, then add the number of Offensive strikes which is also
- significant. Days Lost decreases when there are more Offensive strikes. The
- number of Defensive strikes is not significant, while the total Length is.
- Our final model contains the three variables: number of Strikers, number of
- Offensive strikes, and total Length. We keep the fitted values and residuals
- for a plot.
-
- When we plot Days Lost and Fitted values against any of the independent
- variables, we see that the scale of the ordinate is truncated to 4 digits.
- Thus, we use Constant in the Var Mod menu to scale down (divide) Days Lost by
- a factor of 100 and redo the multiple regression analysis, but this time only
- for the model already found. We can then obtain a suitable plot. The
- regression line is, of course, not straight, since we are projecting a four-
- dimensional space onto two dimensions.
-
- As a second quick analysis of these data, let us consider how the number
- of Strikes each year depends on that of previous years. In the Var Mod menu,
- we create a series of new variables with LagLead, with lags of -1, -2, and
- -3. Regression of number of Strikes on these lagged variables shows that the
- number depends only on that lagged by one year.