Regression models for air pollution and daily mortality: analysis of data from Birmingham, Alabama

Biologically plausible particulate air pollution mortality concentration-response functions.

“…Smith et al (2000a) did not find statistically significant evidence in favor of nonlinear concentrationresponse functions. However, they question whether testing a nonlinear versus a linear concentration-response function is the appropriate formulation.…”

Biologically plausible particulate air pollution mortality concentration-response functions.

“…Smith et al (2000a) did not find statistically significant evidence in favor of nonlinear concentrationresponse functions. However, they question whether testing a nonlinear versus a linear concentration-response function is the appropriate formulation.…”

Robustness of different regression modelling strategies in epidemiology: a time-series analysis of hospital admissions and air pollutants in Lisbon (1999-2004)

“…A MA window of 7 days was selected because: (1) it can give considerable flexibility for varying exposure duration requirements and latent/induction periods; (2) 7 days is still considered an acute time‐scale for associations (WHO, 2005) and (3) MAs of 7 days in both APs and HAs can adjust for biases arising from weekly fluctuations, thus bypassing the need for dummy variables for weekdays. It is also advisable to choose a model specification a priori rather than choose the ‘best’ one by resorting to multiple testing (Lumley and Sheppard, 2000; Smith et al ., 2000).…”

“…First, we have compared statistics from different model specifications without knowing which (if any) is the most accurate; the only criteria we are able to evaluate are the precision, relative change in estimates and the latter's robustness in the face of small changes to the dataset. It must be emphasised that, in the absence of independent experimental evidence to guide the choice of model specification, statistical‐based decisions can lead to strong biases and ultimately meaningless associations (Lumley and Sheppard, 2000; Smith et al ., 2000; Chen et al ., 1999). Secondly, the results presented here refer to the general properties of the 5 model specifications across 72 HA‐AP relationships, and they may differ for specific individual HA‐AP relationships.…”

An Investigation of Distributed Lag Models in the Context of Air Pollution and Mortality Time Series Analysis

“…Previous studies have used DLMs as a measure of PM exposure and suggested that using DLM exposure measures should be preferred to single-or multi-day moving average PM exposure measures. 1,10,15,16 However, this is the first paper to investigate the statistical properties of DLMs as a measure of PM exposure.…”

Biologically plausible particulate air pollution mortality concentration-response functions.

“…Smith et al (2000a) did not find statistically significant evidence in favor of nonlinear concentrationresponse functions. However, they question whether testing a nonlinear versus a linear concentration-response function is the appropriate formulation.…”

Robustness of different regression modelling strategies in epidemiology: a time-series analysis of hospital admissions and air pollutants in Lisbon (1999-2004)

“…A MA window of 7 days was selected because: (1) it can give considerable flexibility for varying exposure duration requirements and latent/induction periods; (2) 7 days is still considered an acute time‐scale for associations (WHO, 2005) and (3) MAs of 7 days in both APs and HAs can adjust for biases arising from weekly fluctuations, thus bypassing the need for dummy variables for weekdays. It is also advisable to choose a model specification a priori rather than choose the ‘best’ one by resorting to multiple testing (Lumley and Sheppard, 2000; Smith et al ., 2000).…”

“…First, we have compared statistics from different model specifications without knowing which (if any) is the most accurate; the only criteria we are able to evaluate are the precision, relative change in estimates and the latter's robustness in the face of small changes to the dataset. It must be emphasised that, in the absence of independent experimental evidence to guide the choice of model specification, statistical‐based decisions can lead to strong biases and ultimately meaningless associations (Lumley and Sheppard, 2000; Smith et al ., 2000; Chen et al ., 1999). Secondly, the results presented here refer to the general properties of the 5 model specifications across 72 HA‐AP relationships, and they may differ for specific individual HA‐AP relationships.…”

An Investigation of Distributed Lag Models in the Context of Air Pollution and Mortality Time Series Analysis

“…Previous studies have used DLMs as a measure of PM exposure and suggested that using DLM exposure measures should be preferred to single-or multi-day moving average PM exposure measures. 1,10,15,16 However, this is the first paper to investigate the statistical properties of DLMs as a measure of PM exposure.…”

Biologically plausible particulate air pollution mortality concentration-response functions.

“…Smith et al (2000a) did not find statistically significant evidence in favor of nonlinear concentrationresponse functions. However, they question whether testing a nonlinear versus a linear concentration-response function is the appropriate formulation.…”

Robustness of different regression modelling strategies in epidemiology: a time-series analysis of hospital admissions and air pollutants in Lisbon (1999-2004)

“…A MA window of 7 days was selected because: (1) it can give considerable flexibility for varying exposure duration requirements and latent/induction periods; (2) 7 days is still considered an acute time‐scale for associations (WHO, 2005) and (3) MAs of 7 days in both APs and HAs can adjust for biases arising from weekly fluctuations, thus bypassing the need for dummy variables for weekdays. It is also advisable to choose a model specification a priori rather than choose the ‘best’ one by resorting to multiple testing (Lumley and Sheppard, 2000; Smith et al ., 2000).…”

“…First, we have compared statistics from different model specifications without knowing which (if any) is the most accurate; the only criteria we are able to evaluate are the precision, relative change in estimates and the latter's robustness in the face of small changes to the dataset. It must be emphasised that, in the absence of independent experimental evidence to guide the choice of model specification, statistical‐based decisions can lead to strong biases and ultimately meaningless associations (Lumley and Sheppard, 2000; Smith et al ., 2000; Chen et al ., 1999). Secondly, the results presented here refer to the general properties of the 5 model specifications across 72 HA‐AP relationships, and they may differ for specific individual HA‐AP relationships.…”

An Investigation of Distributed Lag Models in the Context of Air Pollution and Mortality Time Series Analysis

“…Previous studies have used DLMs as a measure of PM exposure and suggested that using DLM exposure measures should be preferred to single-or multi-day moving average PM exposure measures. 1,10,15,16 However, this is the first paper to investigate the statistical properties of DLMs as a measure of PM exposure.…”