SYNOPTIC ABSTRACT
Consider a times series in simple linear regression. It is shown that under suitable conditions point estimates or predictions for the next time period into the future are unaffected by values of the dependent variable at some given time period in the past. The result is extended to cases other than time series, and to predictions in multiple linear regression. Since under suitable conditions not all the sample data need be gathered for making predictions, considerable cost savings are possible in such practical applications as destructive sampling. This is illustrated with several examples.
Key Words and Phrases: