Mortality Projection: Curve Fitting And Linear Regression
View/ Open
Date
2011Author
Ndekele, Erastus Kimani
Type
ThesisLanguage
enMetadata
Show full item recordAbstract
This study was aimed at predicting future mortality rates given two main scenarios. The first is in the
case where there is a body of mortality tables from the past and the second is in the case where such a
database does not exist.
In the first case, linear regression was used to estimate mortality of specific age groups for a specific
future year. In this case the specified future year was from 2000 to 2009 although similar computations
can be carried out for age specific and single year life tables. To perform these linear regression two
transformations were considered to the existing q type mortality rates: the log linear transformation and
the logit transformation
For the second case, it was assumed that the only data that existed was that of actual deaths and exposed
to risk. Thus, for adequate mortality projections to be done a Makeham curve was fitted in one case and
a cubic spline graduation was done for the second case.
Both the Makeham and the cubic spline methods were observed to have a good fit to the data, the latter
providing a better fit than the former. However the linear regression methods were observed to give an
almost constant difference at every age and would probably not be the best methods of forecasting
especially for very long time periods. The advantage of the linear regression method was that it was seen
to keep the original shape of the graduation
Publisher
School of Mathematics