Consequences of declining survey response rates for smoking prevalence estimates.