Bias in Engineering Estimation
- Randal M. Brush (Arco Oil and Gas Co.) | Sullivan S. Marsden (Stanford U.)
- Document ID
- Society of Petroleum Engineers
- Journal of Petroleum Technology
- Publication Date
- February 1982
- Document Type
- Journal Paper
- 433 - 439
- 1982. Society of Petroleum Engineers
- 1.6 Drilling Operations, 4.1.5 Processing Equipment, 4.5 Offshore Facilities and Subsea Systems, 6.6.2 Environmental and Social Impact Assessments, 7.2.3 Decision-making Processes, 5.7 Reserves Evaluation, 4.3.4 Scale
- 0 in the last 30 days
- 184 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 5.00|
|SPE Non-Member Price:||USD 35.00|
Data from 40 offshore oil and gas production platforms in the Gulf of Mexico have been analyzed for bias. The significant biases found were deemed the result of an optimistic estimation process. Computer modeling shows these biases to have an effect on a company's operations. Several possible causes of and cures for the biases are suggested.
To the layman, petroleum engineering may appear to be an exact discipline. The numerical result of a long series of calculations mistakenly can be regarded as some sort of true value. The engineer, on the other hand, knows that the answers he arrives at are, for the most part, estimates: thus, there is always uncertainty as to their accuracy. What follows is an examination of a large group of estimates (approximately 640) connected with the installation of 40 oil and gas platforms in the Gulf of Mexico. The estimates have been compared with actual values when possible (such as with the time required to install the platform) or with the most recent and presumably best estimate (such as with the total oil reserves recoverable from a platform). Several statistical methods have been applied to the data in an attempt to answer several questions for each of the 16 variables and for the body of data: Are the data biased? What is the magnitude of the bias? Are the biases statistically significant? How do the biases affect the projects' performances? How do they affect the company's performance? What might be the causes of these biases? What can be done to minimize them and their effect on the company? This study is based on an engineer's degree thesis, which in turn was based on data released by ARCO Oil and Gas Co., South Texas Dist., Offshore Section.
Theory and Data Analysis
While much has been written in engineering and management journals regarding the uncertainties present in decision making, little has been reported on the presence of consistent misestimation (bias). Even less has been written on the measurement and interpretation of observed bias. Tversky and Kahneman describe the phenomenon of overestimation and show how biases can be incorporated into intuitive decision processes. They deal with the problem from a decision science point of view. Uman et al. provide a very informative case study of pre- and postdrilling estimates of oil and gas reserves. These estimates were made by the USGS for offshore dulling tracts. While the authors found that the reserves estimates could be considered accurate only to plus or minus an order of magnitude (0.1 x predrilling estimate less than postdrilling determination less than 10 x predrilling estimate), they found no bias in the majority (80%) of data sets (as grouped by Outer Continental Shelf sale numbers). The data examined here were examined concurrently and reported by Franzen et al., and the reader is referred to their paper for additional analyses and conclusions. This study is based on data gathered from Gulf of Mexico production platform files. These data have been used by the Conoco Inc./ARCO Oil and Gas Co./Getty Oil Co./Cities Service Co. consortium in the evaluation and planning of these platforms. Forty platform projects were chosen as the source of the data set. To be included in this study, the project had to have been initiated within the last 15 years and had to be completed or nearly completed by the time the data were gathered. To evaluate these variables for bias, "before" values estimated for the initial project proposal, and "after" values, the actual or most recent estimates of the variables' values, were gathered for each of the 16 variables and for each of the 40 platforms.
|File Size||474 KB||Number of Pages||7|