| Popis: |
* The author wishes to express his appreciation to Alfred D. Basch and Ethel E. Metzendorf of the New York State Department of Commerce and Joseph Lev of the New York State Department of Civil Service for their encouragement and helpful comments. Techniques of overcoming biases in serial correlation coefficients computed from short linear autoregressive time-series are considered. Two alternative estimates of autocorrelation coefficients are suggested as probable improvements over the usual estimates based on averages of observations made over a seasonal period. The alternative estimates are obtained, for example, from 12 yearly series, one for each month, and are: 1. an average of serial coefficients computed from each yearly series; 2. pooled sums of squares and lagged products for each series formed into a single serial correlation coefficient. |
