OPTIMAL RESPONSE SURFACE DESIGNS IN THE PRESENCE OF DISPERSION EFFECTS 69 additional design location would not be necessary. Similar allocations should be made for the second the correct variance ratios.then the tables of optimal should cor pond to the locations of smallest vari designs can be used to identify the appropriate de- sign to us vever,since the exact specincati ations not possible.more general recommendations must be cate as equal a number of runs as possible to each made. cation,while for better prediction allocatea few im that the This will provide atio at thes to the design region of interest ions and will correspond more closely to the op For a first-order or interaction for which timal designs. the standard replicated 22 factorial design is opti References rance exists,ic they are nearly equal) If these design locations are an dd the alls within the c0f0n54.pp.622-5 the list of desigr Whon the the optnaldsgssu possible Cou if prediction ion Effects throughout the design ncs28.p.19.27 re ion is the primary interest. then instead the identified locations where the estimated variance DAVIDIAN.M.and CARROLL.R.J.(1987).Tre where variance is smalle The allocation of runs to the factorial location in hearemothedeig KIEFER. 1.(1959) Optimum Expe First,it makes the design more ba cer e at this location an rticularly in the case of I-optimality,effectively re uces the variance and improv variance structure the replicated 22 factorial desig Experiments is D.and I-optimal for ll variance ratios,and that for the Volumes I and 2 the dratic cated in the ntal De Vol.29.No.1.January 1997 Journal of Quality Technology Reproduced with permission of the copyright owner.Further reproduction prohibited without permission.Reproduced with permission of the copyright owner. Further reproduction prohibited without permission