METHODS
Daily mortality counts are the dependent variables in the
statistical analyses. A Poisson distribution was used, which
assumes that counts of independent, rare events follow a
Poisson distribution, conditional on the explanatory variables.
The dependent variable in the Poisson regression is
the natural logarithm of the expected mortality count,
while the regression coefficients are the natural logarithms
of the rate ratios. The associations between PM10 and six
different mortality end points were examined, including
natural mortality (all deaths net of homicides, suicides,
and accidents), cardiovascular mortality, respiratory mortality,
and natural mortality for several different age groups
including 50 and above, 6–49, and under 6.
To develop our regression model, we determined the
best fit of several covariates before entering PM10 into the
model. This was done to control for factors besides PM10
that vary on a daily basis and that might explain variations
in daily mortality. By entering the PM10 variable last,
we reduce the chance of inadvertently attributing an effect
to PM10 that might have been explained by other factors.
Same-day PM10 exposure and lags up to four days were examined,
along with moving averages of three and five days.
To model the seasonality, we considered several different
approaches, including (1) dummy variables for each
METHODS
Daily mortality counts are the dependent variables in the
statistical analyses. A Poisson distribution was used, which
assumes that counts of independent, rare events follow a
Poisson distribution, conditional on the explanatory variables.
The dependent variable in the Poisson regression is
the natural logarithm of the expected mortality count,
while the regression coefficients are the natural logarithms
of the rate ratios. The associations between PM10 and six
different mortality end points were examined, including
natural mortality (all deaths net of homicides, suicides,
and accidents), cardiovascular mortality, respiratory mortality,
and natural mortality for several different age groups
including 50 and above, 6–49, and under 6.
To develop our regression model, we determined the
best fit of several covariates before entering PM10 into the
model. This was done to control for factors besides PM10
that vary on a daily basis and that might explain variations
in daily mortality. By entering the PM10 variable last,
we reduce the chance of inadvertently attributing an effect
to PM10 that might have been explained by other factors.
Same-day PM10 exposure and lags up to four days were examined,
along with moving averages of three and five days.
To model the seasonality, we considered several different
approaches, including (1) dummy variables for each
การแปล กรุณารอสักครู่..
