3 The microsimulation model

View the most recent version.

This section describes the method used to integrate the cross-sectional analysis presented in the previous section to the microsimulation model to produce projections of future needs of home care services. Assumptions for different projection scenarios are also discussed.

3.1 Research design

The research design for the projection model included both cross-sectional and longitudinal data. The crosssectional aspect of the model was partly derived from the analysis presented in the previous section, using data from the 1996 National Population Health Survey and General Social Survey. The longitudinal aspect of the methods, needed to project the family network, was largely taken from the already existing model developed at Statistics Canada. This model, LifePaths, makes use of many data sets from surveys conducted by Statistics Canada. It allows the user to take into account part of the complexity of the life cycle of individuals who constitute the Canadian population. LifePaths takes account of birth, death, immigration status, interprovincial migration, marital history (including common-law unions), educational history, employment history and the birth and presence of children at home. It is a useful instrument to analyze government policies having a longitudinal component and whose nature requires evaluation at the individual or family level (Wolfson and Rowe, 2004).⁶ Obviously, it is a very useful instrument when studying the changing nature and extent of the family network.

Contrary to more conventional demographic projections using a macro perspective, LifePaths uses individuals as the basic unit of analysis. The microsimulation creates a synthetic cohort of individuals going through their life cycle with different probabilities of having specific events occurring, probabilities that vary across individuals depending on their characteristics. These probabilities are derived using multivariate analysis on various data sets from surveys made by Statistics Canada. Every time an event occurs, e.g. leaving school, probabilities of other event occurring in the future are updated to take into account the new characteristics.

Not every event occurring in someone's life can be studied with longitudinal data. In many cases, the only data available are cross-sectional in nature and cannot provide the information necessary to feed the microsimulation model. For this reason, results presented above were used in conjunction with the microsimulation model to reach our research objectives. Results discussed in the previous section were applied to subsets of the population to derive, for example, the number of elderly persons with a disability or the number using different sources of assistance. Although the microsimulation approach allows an analysis of the life cycle of every individual in the projection, this is not the case if we want to look, for example, at the number of years a person lived with a disability. Of course, this is due to the fact that we did not analyze the process of disability but instead looked at factors associated with having a disability in a cross-sectional survey. However, we do have a count of the projected number of individuals with a disability at different points in time up until 2031. This same approach was used for disability, need for assistance, living arrangement, receipt of assistance and source of assistance. Figure 2 shows how we integrated the longitudinal and crosssectional approach in this research.

The left part of figure 2 lists the characteristics needed in the logistic regressions performed in the crosssectional part of the projections. Of course, the microsimulation produces many more characteristics of each individual, but only the ones presented in figure 2 were used in the logistic regressions. We first had to run the microsimulation to get the population by age, sex, schooling level, region of residence, marital status, age of spouse, place of birth and number of surviving children. We also needed to determine who lived in an institution and who lived in a private household. Only the latter population was used in the logistic regressions as, by definition, home care is provided to those who live in private households. As for the cross-sectional part of the projections, each probability (disability status, need for assistance, living arrangement, receipt of assistance, and source of assistance) was applied in the sequence shown in figure 2. As can be seen, we make the assumption that living arrangement is partly determined by the need for assistance. Living with others is then partially seen as a strategy to cope with a need for assistance related to a disability. This was done only for the non married population, married individuals being assumed to live with their spouse.

Figure 2
The microsimulation model

3.2 Assumptions on future disability trends

Results from the logistic regression showed, as expected, that the main factor associated with the need and use of home care services is the health status of the individuals. Therefore, it seems important to test the sensitivity of the projection model to eventual changes in the health of the population. Three different scenarios were used to answer this concern:

1. Probabilities of disability levels held constant at 1996 levels (constant scenario);
2. Probabilities of disability levels gradually decreasing (compression scenario);
3. Probabilities of disability levels gradually increasing (expansion scenario).

In the case of the compression scenario we assumed that the probability of having a given level of disability according to specific individual characteristics would gradually (over a 15 year-period) go down. This was done by giving to an individual of a certain age the probability of having a specific level of disability of someone 5 years younger. As mentioned, this was done gradually over a 15 year-period. After 15 years (2016), the probability of having a certain level of disability is exactly the one of someone 5 years younger. In the case of the expansion scenario, the approach was exactly the same except that this probability is increased to someone five years older instead of 5 years younger.

Our intention here is not to predict the health of the population in the future, but to analyze the impact of an increase or a decrease in the levels of disability in the future. This allows us to analyze the sensitivity of the model to changes in the health of the population. The next section presents the results of the projections.

 

⁶ . For a detailed description of the LifePaths model: www.statcan.gc.ca.