To provide an arithmetic unit which performs a regression calculation of statistic data, etc., and then easily specify the most approximate regression of totalized data.

Optional list data LIST1 and LIST2 are defined as list data LIST-X and LIST-Y to be calculated respectively and the execution of a regression calculation is designated by the operation of an EXE key. Thus, the coefficients of regression expressions corresponding to nine types of regression (REG1-REG9), for example, are calculated in sequence. At the same time, an output list consisting of Y value obtained by substituting X value of the calculated coefficients and the lists to be calculated respectively is calculated. Furthermore, a residual list is calculated between the input and output lists and stored in a regression arithmetic data memory. Then the sum total (ΣX), etc., is calculated according to the residual list and stored in a result data memory. The result data of each type of regression are compared with each other, and an expression of regression of the type with which the result data of the least value are obtained from the comparisons is displayed and outputted as the regression that is most approximate to the list data to be calculated.