Summary: All trend lines previously described together with Figures 1 through 4, below, are based on ordinary least squares (OLS), without correction for serial correlation nor evaluation of their standard errors of estimate.

As a sequel to this study, a more advanced statistical analysis based on Hildreth-Lu (H-L) non-linear equation solutions, confirmed the absence of serial correlation with the regression residuals. The recent study shows:

(1) Virtual identity for all computed trend line slopes as well as standard errors of estimate between the OLS and H-L procedures.

(2) Virtual absence of serial correlation as judged by the Durbin-Watson statistic together with matching of standard errors of estimate between the OLS and H-L procedures.

(3) Ostensible support, by use of tabulated t-stat values and line-chart patterns, for the conclusion that a preponderance of trend line slopes are statistically or inferentially significant and negative, thereby implying that net annual rates of temperature change are at least definitely not increasing but are more likely decreasing.

The series of four charts displayed below for cities within AZ, GA, MN and VA illustrate a high-degree of order and regularity pertaining to their individual patterns (especially for GA and MN), which argues against randomness.

It is emphasized that these charts are based on OLS. Nevertheless, they closely mimic the results of companion line-charts based on the H-L procedure.

Figure 1. Arizona (click for larger image)

Figure 2. Georgia (click for larger image)

Figure 3. Minnesota (click for larger image)

Figure 4. Virgina (click for larger image)

Table 1, below, specifies the populations for the 53- to 55-year trendlines.

© 2003 by Ask Analytic Services, Inc