Selection of suppliers using Bayesian estimators: a case of concrete ring suppliers to Eurasia Tunnel of Turkey

ABSTRACT

This work introduces a methodology to evaluate, rank, and select suppliers for an organisation managing a large and complex construction project. The company’s procedure to complete a supplier evaluation is conflated with other supplier features such as product type and complexity, delivery characteristics and requirements, and geographic location of the project. The introduced model segregates the effects of each feature and then aids supplier selection on various criteria without the confounding effects. Model parameters are determined using Bayesian estimators allowing for information integration from prior periods. The estimation approach provides rich model parameter data, allowing for use in additional analysis. This work advances the research in supplier selection by illustrating a practical forecasting and predictive technique for supplier selection. One result is that the separability of factors in a multiple criteria decision environment can prove valuable for managers to help decipher and isolate factors in a complex decision environment. The technique is feasible for smaller problem sets and provides a robust solution. Past performance and future performance potential are both considered. Analysis and future research directions allow for further development.

KEYWORDS:

• Supplier selection
• concrete industry
• Bayesian analysis
• civil construction

Acknowledgements

The authors would like to thank ATAS (Avrasya Tüneli Isletme Insaat ve Yatırım A. S.) and Yapi Merkezi Prefabrikasyon A. S., especially Mr. Orhan Manzak, for the invaluable information and insights he provided during this study.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The Jeffreys’ rule states that a prior distribution should be invariant under any monotone transformation of its random variable.

2 $Logit\left(\theta \right)=\ln \left(\theta /(1-\theta )\right),0\le \theta \le 1$.

3 If the precision itself is used as a random variable, instead of variance, then the transformed distribution is called an inverse-gamma distribution. We use a gamma distribution but indicate that the random variable is the reciprocal of variance.

4 The model in this paper has 13 variables $\left(s_i^{\mathrm{\prime }}s\right)$ for supplier effects and 14 variables $\left(p_j^{\mathrm{\prime }}s\right)$ for product effects that are of interest. If all variables are counted, there are 67 variables in the joint posterior distribution in Equation (9).

5 The software WinBUGS is available free of charge from: http://www.mrc-bsu.cam.ac.uk/software/bugs/the-bugs-project-winbugs/, MRC Biostatistics Unit, University of Cambridge.

6 Grand mean of performance scores, $m$, in Equation (1) is 87.52.

7 Information about standard deviation, quantiles, etc. for all revised pdfs are available. We have provided them only when needed to save space and avoid data clutter.

8 The second term, 0.28, is equal to $1/\sqrt{2{\sigma }^2/n}$