THE JOURNAL OF KMUTNB

Volume 32, No. 02, Month APRIL, Year 2022, Pages 415 - 425

A comparison of missing value estimations in randomized complete block design

Chayada Kaewchaicharoenkit, Boonorm Chomtee, Wandee Wanishsakpong

Abstract Download PDF

The study aims to compare the 3 missing value estimations: OLS, MI, and GA in randomized complete block design using both real and simulated data. In the study, there are at least 5% of randomly missing value (m). Two real data sets : t=5 b=3 and t=8 b=4 are used in this research. Also, there are four cases of simulated data : t=3 b=3, t=5 b=3, t=8 b=4 and t=8 b=5. There are 10%, 30% and 50% of C.V., the constant h = 1, 2 and 3. R Program is used to simulate data with 12,000 times for each situation. MSE and MAPE are used as criteria for determination. For the result, when m increases, MSE and MAPE values tend to increase. For the real data, when m=1 and 2, MI has the lowest MSE and MAPE values. For the simulated data, when m=1, OLS has the lowest AVG MSE and AVG MAPE values for almost cases, except when C.V.=10%, h=1, GA has the lowest AVG MSE and AVG MAPE values. When m=2 C.V. = 10% and 30%, it is found that MI and GA have the lowest AVG MSE and AVG MAPE values respectively. When C.V. = 50%, for almost cases, OLS has the lowest AVG MSE and AVG MAPE values.

Keywords

Randomized Complete Block Design; Missing Value; Ordinary Least Square; Multiple Imputation; Genetic Algorithm

---