Further Analyses
Grouping
Analysis
There were three basic district
API trends:
very low API throughout (group 1); high API in the late 1970’s and then
decreasing to below 12 API (group 2); and high API in the late 1970’s,
decreasing in the 1980’s and then increasing again above 12 API in the
1990’s (group 3). Group 1
consisted of Bangalore, Belgaum, Dakshina Kannara, Dharwad, Kodagu, Kolar,
Shimoga, and Uttara Kannara. Group
2 contained Bellary, Bidar, Bijapur, Gulbarga, Mandya, Mysore, Raichur, and
Tumkur. Group 3 consisted of
Chikmagalur, and Chitradurga and Hassan, two districts with the highest API
peaks. I performed regression
analyses of the climatic variables and malaria rates of these groups when their
data was aggregated.
Although many of the linear regressions by group were statistically
significant, they did not produce R2 values that were much higher
than the ones for the individual district analysis.
In most cases, the R2 value was higher than that from the
simple linear regression for all of the districts in each group except one.
For the few cases in which the R2 value was higher due to the
grouping than for the simple linear regression before grouping, it was not
significantly higher and therefore did not contribute much to the analysis.
Removing
Malaria Control Years
When looking at the API trends by district in Karnataka, I had assumed
that the decreases in the late 1970’s and early 1980’s were due to effective
malaria control. By removing these
years from my analysis, I hypothesized that the degree to which climatic factors
predicted malaria rates would increase. I
removed the low API years from API, SPR, average yearly mean temperature,
January mean temperature, April mean temperature, May mean temperature, April
precipitation, November precipitation, and total yearly precipitation only from
those districts that exhibited a large decrease in malaria during the late
1970’s through the 1980’s that I assumed was due to the success of MPO.
However, removing data during years when it appeared that malaria control
was effective and only using years outside of this range in a linear regression
did not improve the R2 values and in many cases these regression
models were not statistically significant.
For
Chitradurga and Hassan, significant correlations were found for the regression
models applied to API and April mean temperature, SPR and April mean
temperature, and SPR and May mean temperature, and SPR and April precipitation.
The R2 values for the removed data were not greater than those
for the regression models without removing data, except for the linear
regression between SPR and May temperature and the one for SPR and April
precipitation which was significantly better.
Without removing data, the R2 value was 0.35, and after
removing data for 1983 through 1992, the R2 value is 0.61.
These were the only instances in which the regression model applied to
the data with some years removed was better than the regression model with all
of the data.
For Chikmagalur, the only statistically significant regression analysis
for the data with years of low API removed was between SPR and April
precipitation, and between SPR and total yearly precipitation.
The R2 values for this analysis was 3.93 times larger than the
R2 value for SPR and total yearly precipitation before removing data
from low API years. For April
precipitation, removing the data from years with low API resulted in an R2
value to be 1.07 times higher than for the original regression.
Although there were a few cases of increased predictive power by climatic
variables when years assumed to be affected by malaria control efforts were
removed, there were many more cases without statistical significance when there
was statistical significance for regressions performed before removing the data.
The years that were removed did not have anomalous temperature or
precipitation trends. It was hard
to know which years to remove, and also the assumption that removed years had
low API because of the effect of malaria control efforts could be incorrect.
If we assume that the decrease in API was due to effective malaria
control efforts, the results from this analysis imply that malaria control
efforts did not decouple the relationship between climate and malaria but rather
that the control efforts just decreased the total prevalence, and that the
malaria was still, to a certain degree, predicted or affected by climatic
variation.
Analysis of Yearly Change in API and SPR
The change in API and SPR, by
district over time, are shown in the following figures, respectively.
These figures highlight the trends shown in the regular API and SPR
trends; net decreases in the late 1970’s, and net increases in the early
1990’s. Some monthly mean
temperature and monthly precipitation trends that did not significantly
correlate with API or SPR, did correlate significantly with the yearly change in
API or SPR, especially February and May precipitation, but the level of
correlation was rather low, never surpassing 0.45, and in most cases, even
lower. For the monthly climate
variables that were significantly correlated with API or SPR in the simple
linear regression analysis, they did not often significantly correlate with
yearly change trends in API and SPR. If
they did significantly correlate, the R2 values were lower.
In general, the change in API or the change in SPR was not found to be a
better variable to use than API or SPR.
