L.2.2 Linear regression

Figure L-3. SAS probit plot of water column toxicity test example data

The various estimates in Table L-5 differed by up to 7.7 percent elutriate,

which is not unusual or alarming. The Probit or Logistic LC50 would be the

preferred estimate, because the regression lines fit the data well, and the

regression methods use more of the data in such cases. However, any of the

estimates would be adequate for use in the mixing model in Appendix H,

because the imprecision and uncertainty involved in the model calculations and

estimates are undoubtedly far greater than the differences among the LC50

estimates.

Acute toxicity endpoints other than 0.01*LC50 can be considered for use in

the mixing model. These endpoints include low toxic effects concentrations

such as LC10 (Moore and Caux 1997; Scholze et al. 2001); the No Observed

Effects Concentration (NOEC) (Capizzi et al. 1985); and the Inhibition

Concentration ICp, where p is a percent reduction from control response (USEPA

1994).

L.2.2 Linear regression

Linear regression may be needed to characterize the site-specific relationship

between suspended solids and turbidity in effluent pathway testing. The

regression equation is used to predict suspended solids concentrations from

turbidity measurements. Linear regression may also be used to calculate the

contaminant distribution coefficient (Kd) in the sequential batch leach test for

leachate evaluation. Kd is the slope of the linear regression of leachate

concentrations versus sediment concentrations of a contaminant of concern for

L25

Appendix L

Statistical Methods