Table 3.

Model structure for the best model (selected by cross-validation) for each pollutant and metropolitan region.

Model	No. of time trendsa	No. of PLS scoresb	df/year in time trendc	Spatial smoothingd
				Long-term average (β0)	Time trend coefficients (βi)
				Baltimore, MD
PM2.5	1	3	4	Yes	No
NO2	1	2	8	Yes	No
NOx	1	2	8	Yes	Yes
LAC	1	3	8	No	No
Chicago, IL
PM2.5	1	3	8	Yes	No
NO2	2	2	4	Yes	Yes
NOx	2	2	8	Yes	No
LAC	1	2	8	Yes	Yes
Los Angeles, CA
PM2.5	2	3	8	Yes	No
NO2	2	3	8	Yes	Yes
NOx	1	3	4	Yes	Yes
LAC	1	2	4	Yes	No
New York, NY
PM2.5	2	3	8	No	No
NO2	2	3	4	Yes	Yes
NOx	2	2	8	No	No
LAC	2	2	4	Yes	No
St. Paul, MN
PM2.5	1	3	4	Yes	No
NO2	1	3	4	Yes	No
NOx	1	3	4	Yes	No
LAC	1	2	8	Yes	No
Winston‑Salem, NC
PM2.5	2	2	4	No	No
NO2	1	3	8	Yes	Yes
NOx	1	2	8	Yes	Yes
LAC	1	2	8	Yes	No
aSelected from either 1 or 2 time trends. bSelected from either 2 or 3 PLS scores; scores were covariates in the mean component of the long-term average (β0) and time trend (βi) fields [denoted by Xi(s) in Equation 3]. cSelected from either 4 or 8 degrees of freedom (df) per year; controls smoothness of estimated time trends. dYes, exponential covariance structure. No, independent covariance structure.