CHAPTER 13 FURTHER APPLICATIONS 13.1 ABOUT THIS CHAPTER Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. There are many areas of application of statistics in business and industry beyond those discussed in earlier chapters. We briefly consider some of these here. We begin with three industries that have traditionally used statistics extensively: the food, beverage, and related industries; the semiconductor industry; and the communications industry. Next, we discuss statistical image analysis – a relatively new field with important medical, security, and other applications. We then provide short descriptions of the role of statistics in various other areas and conclude with a glance into the future. We do not cover all the bases; but the discussion should demonstrate the use of statistics in addressing diverse problems in business and industry (and beyond). Many of our comments are based upon inputs from those working in the areas discussed. 13.2 FOOD, INDUSTRIES 1 BEVERAGE, AND RELATED 13.2.1 The Setting: From Soup to Soap Modern conveniences that “enrich” our lives – such as microwave popcorn, frozen gourmet dinners, and instant chicken soup – would not have been possible without 574 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. scientific advances in chemistry, biochemistry, microbiology, and physics. The food and beverage industry has become a high-technology business. It provides challenging opportunities for the application of statistics in areas ranging from market studies to product development to manufacturing. Similar issues also arise in applications involving many other consumer products, especially the ones that are purchased repeatedly – such as suntan lotion, detergents, light bulbs, bug killer sprays, and paper tissues. 13.2.2 Controlling Product Variability Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Reducing product variability is a major goal in the food, beverage, and related industries. These industries, in fact, provide an ultimate example of the need for robust design. There are differences in the types of applications even within the food and beverage industry, especially between raw foods, such as fresh fruits, vegetables, eggs, and raw meats; and processed foods, such as cereals and soft drinks. Raw Foods Raw foods are often subject to variability in appearance and taste. These differences are often due to variations in nature, especially fluctuations in temperature and rainfall and differences in soil fertility. Such environmental (or noise) variables are difficult or impossible to control. Consumers recognize this and expect – and sometimes even enjoy – some product variability, such as between early season, mid-season and late season fruits. In addition, for most raw foods, the 575 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. purchaser can examine, and sometimes even sample, the product prior to purchase. There are, moreover, various control variables that can be set to make raw foods more homogeneous. These range from the type and amount of fertilizer to apply to when to harvest and market. In addition, the product itself might be controlled by, for example, spraying with varying amounts of preservative. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. The challenges of growing foods more scientifically led to the development of the statistical design of experiments in agriculture during the early part of the last century (Section 2.2.7). These applications tended to focus on improving product yield (e.g., growing the most vegetables or the biggest tomatoes), rather than on minimizing variability. Processed Products Consumers are less willing to accept variability in processed products. In buying chocolate chip cookies, canned soups, bottled beer, and so on, they expect a product to look and taste the same from one purchase to the next. We want cheeses from different packages of a particular brand to taste the same, despite the fact that they come from milk from different cows on different farms at different times. Also, cereal manufacturers need combat the tendency of ingredients with different densities (e.g., nuts, raisins, and grains) to separate during packaging. Similar requirements hold for other repetitively purchased products. Tissues, for example, need to have the same soft feel from one package to the next. Manufacturers strive to achieve product homogeneity for processed products by: 576 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. • Working with suppliers to minimize variability. • Understanding the factors that induce variability during manufacture and addressing these. • Varying processing conditions to adjust for the variability in raw products. For example, to achieve homogeneity in roasted peanuts, measurements on such factors as sugar content and moisture are obtained on crop samples during harvesting. This information is then used to set processing (e.g., roasting) conditions. • Blending product grown at different times or in different places. The processing of flour, for example, involves mixing varieties of soft and hard grains, whose properties vary over time. Homogeneity in the final product is attained by adjusting the mixing proportions, based upon product or process measurements. As for other products, producers need also be concerned with robustness to usage conditions. A dishwasher detergent, for example, needs to be effective in cleaning dishes regardless of water hardness. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.2.3 Data Issues In the food, beverage, and related industries, as in other businesses, data are obtained to help determine what features a product needs to have (and what must be done) to capture the highest possible consumer interest. The ultimate judgment for food and beverage products often involves an assessment of taste and appearance; health considerations are also becoming increasingly significant. Other characteristics may be important for nonfood products; the users of a shampoo might, for example, be concerned with aroma. 577 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Sources of Data Foods, beverages, and related products are typically assessed by: Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. • Consumers (or consumer panels): These generally consist of (potential) purchasers who provide their personal preferences about product characteristics. These are solicited during design and marketing (Section 3.3.1), or they might be on a (possibly prototype) product that has already been made. Consumers describe what they like and dislike about the product, how it compares with competitive products, and gauge their ultimate satisfaction. The results may be used to make improvements and, possibly, to formulate advertising claims (Section 9.8.3). • Instruments: These involve measurements on such product properties as color, texture, and composition to quantify chemical or physical characteristics. • Experts (or panels of experts): These typically are trained employees who make sensory assessments of product properties that cannot be measured by instruments (e.g., juiciness, puffiness, and aroma), generally without making value judgments. A typical (and highly envied) example is that of wine tasters. Such evaluations are most often made on final product as a quality check. A common (simplified) sequence of activities is displayed in Figure 13.1. Types of Data Measurements obtained by instruments are, by and large, similar to those for other products. In contrast, evaluations by consumers tend to be more subjective and are often expressed on an ordinal (intensity or acceptance) scale, such as from 1 to 5. Such numbers may express preferences running from completely unimportant to highly important or assessments of the product as poor, fair, good, very good, and outstanding. Relative product rankings may be used for assessing 578 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. competitive products. Sensory assessments by experts, such as an assessment of the degree of dryness of a wine or softness of a tissue, may also be on such a scale. Figure 13.1 Data sources in life cycle of food products. Data Gathering As in other applications, the data-gathering process is highly important. We generally strive to obtain a random sample of product, and to come as close as possible to a random sample of consumer evaluators.2 Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. When many samples are to be evaluated in the same session, protocols that minimize the impact of the sequence of evaluation need to be developed; this is especially important in taste tests.3 The resulting sensory and preference data are inherently variable; there are differences in how individuals perceive product attributes, how consumers and testers use scoring scales, and their stamina over multiple samples. Benchmarks for consistently assigning scores need to be established. GRR studies (Section 3.4.2) might be conducted to assess measurement consistency. Data Analysis As we have seen, the data resulting from consumer preference and expert sensory studies are often ordinal (e.g., intensity scale from 1 to 5) or categorical (e.g., fruity, acidity, and bitter taste). Statistical analysis of 579 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. such data requires special methods (see Agresti 2002, 2007; Hildebrand, Laing, and Rosenthal 1977). (Also see Lea, Naes, and Rodbotten 1997 for a discussion of specialized methods for the analysis of sensory data.) 13.2.4 Product Formulation: Mixture Problems As a Southern general wrote, mixing mint juleps “is a rite that must not be entrusted to a novice, a statistician, nor a Yankee.” —W. Faulkner Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Statistically planned experiments are frequently used to help develop the best possible food, beverage or related product, based upon consumer preferences, manufacturing efficiency, cost considerations – or, most likely, a combination (see Hare 2006). These typically call for determining an optimum product formulation, as well as optimum processing conditions. Processed foods generally involve mixtures of ingredients. For a baked product, for example, we may wish to determine the best blend or mix of ingredients (e.g., yeast, flour, water, shortening, eggs, salt, and spices). This suggests the use of a specialized type of designed experiment, known as a mixture experiment – so-called because it involves a mixture of product ingredients that is constrained to add to 100%. Cornell (2002) describes a mixture experiment to identify the optimal blend of three fish species (mullet, sheephead, and croaker) and the best settings of the process factors (duration of deep fat frying, 580 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. cooking temperature, and time) to make fish patties with consistent texture. See Cornell (2002) and Smith (2005) for detailed discussions of the planning and analysis of mixture experiments. 13.2.5 Manufacturing The manufacture of processed foods involves such operations as mixing, storing, drying, condensation, and heat treatment. These operations are similar to those encountered for many continuous processes and considerations similar to those discussed in earlier chapters apply. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.2.6 Ensuring Food Safety The Basic Concern Ensuring a safe product is a key concern to the food and beverage industry. Potential hazards can be biological (e.g., a microbe), chemical (e.g., a toxin), or physical, such as glass or metal fragments implanted in the food.4 The Traditional Approach to Ensuring Safety Industry and regulators have traditionally depended on spot-checks of manufacturing and random sampling of final product to help ensure food safety. This approach, though necessary, tends to be reactive and often fails to identify problems and their causes speedily. A Proactive Approach The U.S. Food and Drug Administration and Department of Agriculture are 581 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. promoting a preventive approach to food safety, known as Hazard Analysis and Critical Control Point (HACCP) similar to the proactive approach for product design, described in Chapter 3, “for use in all segments of the food industry from growing, harvesting, processing, manufacturing, distributing, and merchandising to preparing food for consumption.”5 Issues addressed range from analyzing “potential safety and health hazards” to establishing “recordkeeping that ensures good data are consistently obtained and maintained on potential hazards and their causes.” (See Surak 2007 for further information.) Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.2.7 Food Spoilage Spoilage of a product over time is a further concern of the food and beverage industry, especially for such products as milk, cheese, and meats. There are again strict guidelines and standards to which the industry need adhere. Many foods are labeled with a date by which the product should be sold or used. The determination of such dates raises issues similar to those for pharmaceuticals (Section 10.4.4). 13.3 SEMICONDUCTOR INDUSTRY 6 13.3.1 The Setting Semiconductors, by controlling the flow of electrical signals, are the engine behind much of today’s advanced electronic technologies (e.g., computers, cell phones, digital televisions, airplane control systems, etc). They may be in the form of discrete devices, such as individual 582 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. transistors, resistors, capacitors and diodes. Or they may be connected together in large numbers on a continuous substrate as an integrated circuit (IC). The basic element is a chip, built in a tiny rectangle on a thin wafer sawed from a cylindrical ingot of extremely pure, crystalline silicon. A chip may have millions of transistors. The dimensions are, moreover, getting continuously smaller – even as the specifications on properties are getting tighter. A single wafer can yield hundreds, thousands, or even tens of thousands of chips. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.3.2 Business Features The Manufacturing Process Semiconductor manufacturing takes place in clean rooms and involves complex processes, starting with the fabrication of a seed crystal. This is sliced into wafers that may range in size from a small chocolate chip cookie to a large pizza. The processing of wafers typically requires many steps; several wafers are often worked together in lots. The wafers are eventually diced and fabricated into ICs or discrete devices. Process Measurements Automated measurements are taken during and after each step, or after several steps, of the fabrication process on all or selected wafers and chips and on the generally expensive tools doing the processing. More intensive measurements might be taken when the process is deemed “unhealthy.” Measurements may be used for early-stage rejection (to avoid the cost of further processing of inferior material) or rework or for adaptive control (see Sidebar 13.1). Further measurements, often 583 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. based on machine sensor trace data, are used for fault detection, and still others for control charting (Section 6.7) and for providing information for improving the process. SIDEBAR 13.1: INTEGRATING ENGINEERING PROCESS CONTROL (EPC) AND SPC IN THE SEMICONDUCTOR INDUSTRY Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. The integration of EPC and SPC (Section 8.2) in semiconductor fabrication may involve using the results from one stage of wafer processing to automatically make changes in the process settings in subsequent steps. This can get a drifting process back on target and/or reduce variability, while at the same time allowing the variability to be monitored by control charts. Sachs, Hu, and Ingolfsson (1995) describe an application aimed a keeping the thickness of epitaxy layers on wafers on target. End-of-Line Measurements A series of automated measurements on electrical properties, such as on current and threshold voltage, are typically taken on each chip at the end of the process. The chip is accepted if all measurements meet specifications (see Sidebar 13.2). SIDEBAR 13.2: END-OF-LINE YIELD 584 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. End-of-line yield is the percentage of chips that meets final product specifications on all measurements. End-of-line yield is often low for new products; yields of less than 50% are not uncommon and may be initially acceptable. Due to the high volume of product and the low cost of inspection, economic analysis may suggest 100% end-of-line inspection to control quality of early product – while striving to improve the process upstream. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Over time, end-of-line yield needs to improve dramatically. The ability of an operation to raise its yield rapidly is a hallmark of its success and involves all stages of production. Mature processes typically and consistently yield over 90% acceptable product. The accepted chips may be sorted into groups, based upon their gigahertz rating and possibly other properties. It is important that the process yields a high percentage of “fastrunning” chips; these command the highest price. If a chip’s electrical characteristics do not fully meet prime product requirements, it may, however, still be useable as a less-demanding product. Complex statistical yield models have been developed. Articles on this subject appear in the IBM Journal of Research and Development and in 585 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Semiconductor International, including a series of papers by Atchison and Ross (1999). A Data-Intensive Business Quoting Czitrom and Spagon (1997), “the chemistry and physics of these (semiconductor) processes are not well understood.... Semiconductor manufacturing is thus a measurement intensive business.” Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Statistical methods harness the data to improve each of the many process steps and tie them together. They have also helped shift industry thinking from the compilation of yield rates to understanding underlying causes and how to address them. Data-mining methods are used to find patterns of defects traceable to a particular process step; statistically designed experiments and simulation are also increasingly employed. These are integrated with whatever physical models are available. 13.3.3 Understanding the Impact of Process Variables upon Performance A key industry goal is to improve productivity and yield across the board, and to establish optimum process specifications. It is, therefore, critical to gain an understanding of how the many factors associated with processing (e.g., photolithography, etching, stripping, diffusion, ion implantation, and deposition) impact electrical performance and yield. 586 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Such understanding is sometimes gained, especially during early stages of development, through laboratory-scale investigations. Frequently, however, studies are conducted directly on a (pilot) production line with good units sold to customers. A manufacturer might, for example, make a variety of design masks to determine which provides the best results. Experiments may focus on a specific process step or several such steps. Such studies raise various technical challenges (see Sidebar 13.3). SIDEBAR 13.3: CHALLENGES SOME TECHNICAL Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Challenges in studies to understand the impact of process variables upon semiconductor performance include: • The frequent need to consider a large number of potential impacting variables, especially in experiments involving multiple processing stages. This typically results in the intitial use of screening experiments (such as fractional factorial plans) to efficiently identify a smaller set of critical factors. • Interactions between variables in different steps of the process – for example, between implant and anneal. • The multivariate nature of such studies – due to multiple, and often correlated, response variables. (A response variable at one stage might be a process variable in a subsequent stage.) • The relatively long elapsed time to conduct even a single production run. The fabrication process often spans weeks and even months. 587 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. • The constant pressure to upgrade, calibrate, and maintain state-of-the-art metrology instruments capable of measuring nanometer features with accuracy and precision. (See Baron, Takken, Yashchin, and Lanzerotti 2004 for further discussion and references.) 13.3.4 Monitoring and Controlling Critical Process Variables Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Control charts, especially and range charts, are often used to monitor critical parameters and to detect changes throughout the manufacturing process. These help decide when corrective action needs to be taken to ensure the process stays on target and to reduce variability. Understanding the structure of the process is critical to establishing the most effective groupings for this purpose; wafers or sometimes locations on a wafer are often used. Due to the numerous operations involved in semiconductor fabrication and the many measurements taken, the number of such control charts can be quite large. Therefore, it is useful to develop an automated approach for assessing the results and providing appropriate notifications for out-of-control situations, while striving to minimize the number of “false alarms.”7 13.3.5 Assessing Product Life Accelerated life tests (Section 5.7.8), especially at high-voltage stress, are frequently used to assess the life of 588 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. ICs, as well as that of printed circuit boards, and larger assemblies. Observed failure mechanisms need to be understood and modeled both physically and statistically to ensure they do not have a deleterious impact on usable product life (see Li, Christiansen, Gill, Sullivan, Yashchin, and Filippi 2006; Li, Yashchin, Christiansen, Gill, Filippi, and Sullivan 2006). 13.3.6 Product Burn-In Newly designed ICs are frequently subjected to burn-in (Section 5.11) prior to shipment to minimize infant mortalities in the field. Such burn-in might be discontinued as the product matures. Statistical concepts help determine how long units should be burned in, based upon cost considerations (see Kim and Kuo 1998). Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.3.7 Sources of Further Information Czitrom (2003) provides a detailed discussion of statistical approaches and methods dealing with process improvement for the semiconductor industry (see Sidebar 13.4) and Czitrom and Spagon (1997) describe 24 statistical case studies. SIDEBAR 13.4: STATISTICAL METHODS USED IN SEMICONDUCTOR INDUSTRY 589 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Many of the statistical methods Czitrom (2003) describes have already been mentioned in this and earlier chapters. These include: • GRR studies to quantify measurement variability and act thereon. • Multi-vari charts to compare characteristics over different positions on a wafer, different wafers and different manufacturing lots. • Variance component analysis to quantify the relative contribution to total variability of, say, lots, wafers, positions within a wafer, individual units, and measurement error. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Other methods tend to be unique to the semiconductor industry. For example, wafer maps provide a graphical comparison of a measured performance characteristic by wafer site. Also, the (NIST/Sematech) online Engineering Statistics Handbook contains many case studies, several from the semiconductor industry. 13.4 COMMUNICATIONS INDUSTRY8 13.4.1 Setting the Stage The communications industry is concerned with the transmission of voice, data, text, sound, and video. This industry has, almost since the advent of the telephone in the early twentieth century, played an important role in the development of statistical applications and concepts. As indicated earlier, the study of processes through statistical 590 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. monitoring can be traced back to Shewhart’s work at Bell Labs. Since then, the industry has benefited from the involvement of many key figures in statistics. This attention is not surprising in light of the massive amounts of data generated every minute of every day. Each communication, such as a telephone call, can produce data on who initiated the communication, who received it, where and when it was placed, how long it lasted, how it was paid for, and so on. In addition, there is extensive systems information, such as on switching performance. All of this creates formidable challenges in recording and maintaining data and using such data to the advantages of the business.9 Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.4.2 Usage Assessments Telephone usage data are explored (or mined) by network engineers and statisticians to identify meaningful patterns and relationships, both short-term and long-term. Such studies help address practical questions such as determining: • The capabilities needed to allow calls to be successfully placed and completed 99.99% of the time. • Peak usage times (e.g., 10:00 a.m., U.S. Eastern Standard Time, on Mother’s Day) in the United States and the probability of successfully making a connection on the first try at such times. • A company’s penetration into different geographical markets. 591 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. • The impact of new technologies, such as Voice Over Internet Protocol (VOIP), on customers’ calling habits. See Raftery, Tanner and Wells (2001) for further discussion and added applications. Also, Lambert and Pinheiro (2001) studied wireless calling records for 96,000 customers, making about 18 million calls in three months. Their work characterized statistically the calling patterns of individual customers with the ultimate goal of identifying unusual events that might signal fraudulent calls – a topic considered directly by Cahill, Chen, Lambert, Pinheiro, and Sun (2002). Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.4.3 Reliability and Service Improvement The communications industry, like other businesses, strives for high reliability and service. The complexity of communications systems in general, and phone networks in particular, make them subject to hardware failures or possible performance degradation over time resulting in deteriorating message quality. The reasons for failures during operations need to be quickly identified and addressed. Software issues can, in addition, lead to service breakdowns, potentially involving large geographic areas; software reliability (Section 5.12) is, therefore, of particular interest. Storms are a further major reason for disruption in service of communications products and are best addressed proactively by making systems more robust to adverse weather. Service providers also strive to predict 592 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. emergencies and to develop and evaluate the effectiveness of alternative contingency plans. 13.4.4 Business Assessments Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Pricing The communications industry is highly competitive. The cost of making calls is very important in gaining market share. Service providers need to know the impact of alternative pricing strategies and may test these out in selected markets. This leads to such statistical issues as deciding which market areas to select, the duration of tests, and the selection, and relevance, of the prior (and possibly subsequent) time periods and market areas with which comparisons are to be made. Competitive Evaluations Communications service providers – and especially those of wireless products – often wish to make claims about the superiority of their product’ s performance. Becker, Clark, and Lambert (1998) describe a study that involved hundreds of millions of test calls to quantify reliability for different circuit-based service providers. The study required placing calls on carefully selected test phones between designated cities, say every 15 minutes, using a company’ s own network and those of competitors and recording fraction of calls blocked. Logistics Statistical and related concepts are applied to address various logistical issues, such as optimizing cell tower placement. 13.4.5 Internet Opportunities 593 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. The Internet has had a major impact on the communications industry. Many companies have expanded their services to become Internet providers to private and commercial customers. Statistical methods are being used to gain a better understanding of Internet traffic patterns to help improve performance. Also, companies wish to leverage data from Internet business transactions to learn about customer behavior. Online retailers, for example, may personalize their offerings, or their advertising, by tailoring them to individual customer groups, using, perhaps, multivariate statistical analysis methods, such as cluster analyses, to establish such groups. 13.5 STATISTICAL IMAGE ANALYSIS FOR MEDICAL, SECURITY, AND OTHER APPLICATIONS 10 Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.5.1 Introduction The digital image – simply a matrix of numeric pixels (picture elements) represented as brightness on a computer screen or as a printout – is the natural data output format for a wide variety of applications. From digital cameras and flatbed scanners to medical imaging devices, digital images are being produced ubiquitously in staggering volumes. In fact, in many imaging domains so much data are being produced that it has become nearly impossible for humans to review it all. State-of-the-art computed (axial) tomography (CAT) scanners can, for example, generate a thousand images of a person’s lungs in just a few seconds. This amount of information can be 594 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. overwhelming for radiologists looking for lung cancer, especially when they must review scans from dozens of patients each day. Automatic image analysis holds the promise of sifting through large amounts of data to identify, quantify, and summarize the most important characteristics. Systems are already in use that automatically read addresses on envelopes, detect breast cancer in mammograms, and match crime scene fingerprints to databases of known felons. Many of these applications require modeling uncertainty in the data and therefore rely, to a significant degree, on statistical concepts and analysis. This section briefly describes two of the most important and challenging areas of statistical image analysis: image restoration and pattern recognition. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.5.2 Image Restoration Goal The goal of image restoration is to take an image that demonstrates some defect, such as uneven lighting or being out-of-focus, and alters it to remove the defect. The statistical element of this problem arises from the need to model how an image with a defect can arise when using a particular imaging device such as a digital camera. Although we might be able to model the mechanics of the camera almost deterministically, we are usually uncertain as to exactly how it was used and in what environment. Statistical models can help quantify and address such uncertainty. 595 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Application to Digital Photography One of the advantages of digital photography over conventional film is its ability to correct small “errors” by the photographer. This is accomplished by using one of various easy-to-use and powerful photo-editing software packages. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Many of the image-editing tools in these packages can be considered to be implementations of statistical image analysis. For example, one of the most widely used tools, the UnSharp Mask (USM) filter, addresses the familiar blur seen in out-of-focus photographs caused by the optics of the camera when the focus distance is slightly off the subject of interest. The USM filter recovers sharpness in such pictures. This filter is based on the notion that an out-of-focus picture can be modeled as a Gaussian blur of the desired (but unobserved) in-focus picture. A Gaussian blur is simply a weighted moving average over all the pixels in the image, where the weights are defined by a Gaussian distribution.11 The model assumes that at each pixel of the in-focus image, the blurred image is made up of the weighted average of nearby pixels, weighted more heavily for close-by pixels. To recover sharpness from the out-of-focus image, the USM filter estimates the parts of the image lost due to the camera being out of focus by subtracting a slightly Gaussian blurred version of the out-of-focus image from the original. This difference image, which accentuates the sharp portions of the image (those areas with adjacent structures of large differential brightness), is then added to the original out-of-focus image to increase sharpness and to produce an estimate of the in-focus image. 596 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Simple Example Figure 13.2 gives a stylized example of the Gaussian blur model. In the left panel we have drawn a simple happy face in black on a white background. The middle panel is a Gaussian blur of this image. The goal in image sharpening is to estimate the original in-focus image from this out-of-focus image. The right panel is the result of this estimation process using the USM filter. Like all statistical estimation procedures, it is not a perfect reconstruction. For example, the separation between the left eye and the left side of the face, which is clear in the original image, is less defined in the estimated image. And, all the lines in the estimated image are thicker than in the original. Such artifacts are unavoidable because of the limited information contained in the middle image, which was the input to the filter. However, the contrast in the original image has been largely recovered and, overall, the sharpened estimated image tends to be more pleasing to the eye than the out-of focus image. Photography Example Figures 13.3 and 13.4 illustrate this process in a photography setting. The original slightly out-of-focus image, shown in the left panel of Figure 13.3, was taken using a digital camera. The sharpened image is shown in the right panel. It provides greater contrast between adjacent structures with differential brightness as compared to the original image. For example, in the sharpened image, the reflected light in the boy’s eyes is more noticeable, and the patch on his coat is better defined. Figure 13.4 highlights these differences by zooming in on the boy’s eyes and the patch on his coat. Figure 13.2 A simple example of a Gaussian blur model and the reconstruction of the original image, using a USM 597 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. filter, showing (a) an in-focus image, (b) its Gaussian blur, and (c) the restored image. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Figure 13.3 Original slightly out-of-focus image (left) and sharpened image (right). Figure 13.4 Two portions of the image in Figure 13.3, showing the original slightly out-of-focus image (left) and the sharpened image (right). The USM filter gives the appearance of sharpness to blurry images by increasing contrast between light and dark 598 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. adjacent structures. It cannot recover detail lost due to a camera being completely out of focus. This is because the only input to the filter is the out-of-focus image; any fine structure not actually present in an image cannot be recovered. 13.5.3 Pattern Recognition Goal Image pattern recognition is used to determine if a pattern of features is present in an image, and if so, to quantify the location, size, and shape of the pattern. This has numerous important medical applications, such as in detecting tumors. We will, however, discuss an application dealing with photographic face recognition. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. There are various components of uncertainty in such problems, including whether or not a particular feature is present in the image, biological variability in the locations of features, and noise in the image data itself. Face Recognition Automatic face recognition has become an important tool in criminal investigations and in various security settings. Photographs of a suspected criminal are matched with an often large database of pictures of known persons. The resulting system takes a digital image of a suspect’ s face and analyzes it to identify and characterize various “biometric features” of that person. Typical features are the locations of salient points on the face involving a person’s eyes, mouth, nose and chin. The system then scans a database of known persons searching for matches with individuals whose images display similar features. 599 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Figure 13.5 illustrates several biometric features of an image for a particular person (i.e., the centers and corners of his eyes, corners of the mouth, and the tip and underside of the nose, overlaid as white squares.) At least two parts of this problem rely on statistical analysis. First, we must automatically locate the positions of the biometric features in the given image. Second, we must determine whether the arrangement of these features appears to be consistent with those of any specific image(s) in an available database. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Figure 13.5 Photograph with several biometric feature estimates overlaid. 600 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Locating Biometric Features One method of locating biometric features in an image is to apply a template of features (the white squares in Figure 13.5) to the image and adjust it (via scaling, rotating, and warping) until the best template location is found. The best location is typically defined as the one which deforms the template the least while maximizing some image metric on the pixels underneath each of the template points. This metric could, for instance, be the correlation of the pixels in a small neighborhood around the feature position with those of a gold-standard feature image. Such a gold-standard image could be defined a priori to be an average image estimated from the database or possibly an external image whose biometric features had been defined by extensive physical analysis. Matching with Database Constructing an effective face recognition system calls for matching the biometric features of an unidentified person, such as the one shown in Figure 13.5, with those of an identified person (or persons) in the database. To do this, natural variability in biometric features of different individuals must be properly modeled. For example, the biometric features representing the inner corners of the left and right eye are almost horizontal to each other for most persons (in properly posed photos), but the distance between them varies appreciably from one person to the next. To capture such variability, a multivariate statistical model to represent feature locations can be fitted, using the data in the image database. This model is then used to derive metrics of template variability (sometimes called distance) of the biometric features for the person under scrutiny 601 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. from those for each person in the database. Given a new image to be matched, we first locate its biometric features, as outlined above. Then we decide which photos, if any, in the database match this new image. This is done by using the statistical model to calculate the distance from the features detected in the image under scrutiny to each of the image feature sets in the image database. The persons in the database whose images have the smallest distances are determined to be the best matches to the person under scrutiny. We could also establish a system that flags those matches whose distances are smaller than a specified decision threshold value. This requires trading sensitivity against specificity in the matching algorithm. High specificity means that if a true match does not exist, then the algorithm will correctly fail to come up with one. High sensitivity means that if there is a true match in the database, the algorithm will likely find it. These competing requirements resemble the balancing act between Type I and Type II errors for a simple statistical hypothesis test. The preceding method might be enhanced by using a Bayesian approach, as described briefly in Sidebar 13.5. SIDEBAR 13.5: BAYESIAN METHODS FOR PATTERN RECOGNITION In using a Bayesian approach for pattern recognition we specify, for each image in the 602 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. database, a prior probability of a match to the image under scrutiny – before considering any information about the image, as gleaned from the database. Thus, if we had added information on subjects in the database (beyond their images) – for instance on age, gender, or date of birth – we would use such information to assign higher prior probabilities to individuals whose characteristics seem more likely to match those of the person under scrutiny. If, on the other hand, we have no such prior information, we might say each image in the database is equally likely to match the image under scrutiny and make each of the prior probabilities equal. Using a statistical model we would then, for each image in the database, calculate the probability (or likelihood) of observing the biometric features of the subject under scrutiny under the assumption that the two subjects are the same. Using Bayes’ Rule, the posterior probability that the two images actually match is proportional to the product of the likelihood and the prior probability. This posterior probability is then calculated for each image in the database, and those with the highest posterior probabilities are identified as the most likely matches to the person to be identified. 13.5.4 Further Information 603 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. McCulloch, Laading, Wilson, and Johnson (1996) give a detailed face recognition example. See Duda, Hart, and Stork (2000) for a more general discussion of pattern recognition, and Winkler (2006) for other examples of image analysis discussed from a statistical perspective. 13.6 OTHER APPLICATION AREAS 13.6.1 Health Care Patient Care There is a continuing need worldwide to improve patient care. Statistics plays an important role in such areas as the: Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. • Comparison of the effectiveness of alternative treatments of a disease and of different approaches to patient care, such as, for example, the length of hospital stays after childbirth. • Assessment of waiting times in doctors’ offices or in hospitals and identification of their likely causes. • Evaluation and reduction of errors in filling prescriptions, administering medications, performing surgeries, or in diagnoses. Occasionally, such applications call for conducting a designed experiment. Most of the time, however, experimentation is inappropriate and we must, instead, rely on data from observational studies. As always, planning to get good data from such studies is essential. Process Improvement Health care organizations, due to their highly process oriented nature, present important opportunities for improvement similar to those for businesses. 604 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. The scheduling of doctors in a hospital emergency room provides a typical example. How many doctors are likely to be needed at any given time is predicted from past data on incoming patient flow. Doctors are then scheduled, taking into account their preferences and the need to avoid fatigue. Scheduling outpatient procedures and physical resource planning (e.g., assignment of scanning equipment) are some additional applications. Further Reading Health care practitioners have come to use quality improvement methodologies, such as PDSA, as well as statistical and operations research methods. Relevant articles, for example, Van Den Heuvel, Does, and Bisgaard (2005), appear in Six Sigma Forum Magazine and also Quality Progress. At a more technical level, Woodall (2006) presents a detailed review of issues that arise in the application of control charts to health care. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.6.2 Epidemiological Studies Epidemiology is the branch of medicine that deals with the study of the causes, distribution, and control of disease in populations. Examples of important issues that are addressed using a statistical approach are: • Identification of an impending epidemic by, for example, studying occurrence rates in different regions of a country or continent, or the entire world, over time. • Developing hypotheses concerning likely causes of diseases by relating these to health outcomes. Such evaluations may assess the impact of environmental conditions by geographical region, for example, correlating air quality to cancer incidence in different localities. Or they may study 605 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. personal habits and be conducted on individuals, such as relating the incidence of cancer to smoking. • Assessing the impact of different preventative strategies, such as publicity about methods for prevention, on occurrence of a particular disease. Evaluations such as the preceding rely heavily on observational studies. They also provide important opportunities for using a proactive approach for disease prevention. (See Kahn and Sempos 1989 for a discussion of statistical methods in epidemiology, and Ahrens and Pigeot 2007 and Kleinbaum, Sullivan, and Barker 2007 for more general discussions.) 13.6.3 National Security Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Banks (2005) considers the application of statistics to national defense and homeland security. He describes eight opportunity areas such as: • Probabilistic risk assessment: This balances cost of investment against risk reduction. Statistics has long been used in this area, dating back to studies of the safety of nuclear reactors. • Cybersecurity: This aims for early detection of sabotage directed at defense networks, financial services, public utilities, etc. Automated statistical algorithms are used to detect system anomalies quickly and adopt appropriate countermeasures. • Privacy protection: This includes determining how much noise to add to data to protect personal identity, and other ways of balancing the needs for security and confidentiality. 606 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Banks’ main point, however, is that, at the time of writing, the statistics profession in the United States was not yet sufficiently involved in such activities. 13.6.4 Space Exploration Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Space exploration presents special challenges since again human life may be at stake. Ensuring high reliability prior to launch, especially for “manned” space vehicles, is critical. Improper analysis of the available data on the susceptibility to failure of O-rings in cold weather led to failure to postpone the fatal 1986 Challenger space shuttle launch (see Dalal, Fowlkes, and Hoadley 1989). Statistics is also used to assess the risk caused by external factors, such as ice and insulation foam striking a shuttle during launch, for addressing the uncertainty in the physical models that predict where a launched space vehicle will be at a particular point in time, and for many other applications. Also, Rhew and Parker (2007) describe a study in which the design of experiments was used to “efficiently identify and rank the primary contributors to the integrated drag over the vehicle’s ascent trajectory” for the launch alert system of the NASA Crew Exploratory Vehicle. 13.6.5 Oil Exploration Exploration for oil (and for valuable minerals) may provide the ultimate application of decision-making that balances statistical Type I errors (claiming that oil is present when it is not) and Type II errors (failing to detect oil if it is there). Moreover, if it seems likely that there is 607 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. oil, we need to estimate the quantity and exact location. These assessments typically involve the use of relationships established from past data between physically relevant predictor variables and oil discovery. Such data are of varying relevance and precision; thus, the evaluations might involve using an appropriate weighting scheme and/or Bayesian methods. (See Agterberg and Bonham-Carter 1999 and Harbaugh, Doveton and Davis 1977 for further discussion.) Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.6.6 Environmental Studies Environmental concerns are becoming increasingly important to business and industry. Issues may arise as a result of the manufacturing process itself, such as the discharge of undesirable liquids into the ground or river. Or they may be the consequence of the use of products, such as automobiles or locomotives that emit pollutants. Traditionally, statistics was used to address questions such as “does the discharge exceed the ‘allowable’ limits established by the company, the industry or a regulating authority?” Sometimes, it was also desired to determine whether “hot spots” or highly contaminated areas were present and, if so, to identify their locations. Over time, the focus has shifted to a more proactive approach; that is, to minimize discharges and to find effective pollutant control methods. Statistical methods are typically used in environmental studies to: 608 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. • Model and monitor (using control charts) pollution at a particular location over time.12 • Evaluate seasonal patterns and long-term trends in pollution levels using time series analysis. • Assess the precision of the instruments used to quantify pollution using GRR studies. • Characterize the geographical distribution of environmental pollution using spatial methods (see Cressie 1993). Environmental studies often involve analysis of censored data due to the pollutant level being below (left-censored) or sometimes above (right-censored) the capability of the measurement instrument. (See Gibbons and Coleman 2001, and Gilbert 1987 for further discussion.) Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.6.7 Chemometrics Chemometrics is a specialized area that deals with the analysis and modeling of chemical data. A common problem is identifying the composition of an unknown material based on the substance’s spectra (e.g., reflectance at various wavelengths). Modern instrumentation often generates large amounts of spectral data – a million or more data points per sample are not atypical. Such data are then used to compare the spectra of the unknown material with those of known materials in order to determine its composition. Readily available software packages allow chemometric studies to be carried out directly by a laboratory chemist. These typically include multivariate statistical analysis and 609 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. modeling methods. (See Beebe, Pell, and Seasholtz 1998 for more information.) 13.7 EMERGING AREAS: A GLANCE INTO THE FUTURE We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten. —Bill Gates Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. What are some of the new areas in business and industry in which statistics will be important in the future? We briefly consider this question, carrying forward our discussion in Section 2.4 and elsewhere. (For more detailed assessments, see Lindsay, Kettenring, and Siegmund 2003, 2004; Raftery, Tanner, and Wells 2001; Steinberg 2008; Straf 2003). 13.7.1 Advances in Information Science Computers in the future may weigh no more than 1.5 tons. —Popular Mechanics, 1949 The interplay between information science and statistics was discussed briefly in Section 2.4.1 (see also Wegman, Said, Scott, and Solka 2009). Advances in data gathering and in computing continue to lead to new areas of application, as well as new statistical methods. 610 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Data Gathering The new opportunities presented by the continuing advances in data-capturing capabilities and increased access to databases have been highlighted throughout this book. We have stressed the importance of careful early planning of the data-gathering process to get good data for statistical analysis. This will continue to be critical as added capabilities for using data emerge. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Methods currently in use for tracking tagged units, ranging from product inventory to cattle, are, for example, being extended to such diverse applications as product usage assessment, customer behavior evaluation, fraud detection, field monitoring of products and evaluation of Internet traffic. The advent of radio frequency identification devices (RFID), in particular, calls for “an essential data management infrastructure” to effectively utilize the resulting “data avalanche” (quote from RFID Study Group at Pennsylvania State University 2006). Computations Advances in computer capabilities and in the speed and ease of calculations continue to revolutionize the way statistics is applied in addressing problems in business and industry. The following are just a few examples of such computer-intensive methods: • Modern Bayesian methods (see Gelman, Carlin, Stern, and Rubin 2003). • Markov Chain Monte Carlo simulation methods (see Kass, Carlin, Gelman, and Neal 1998). • Experimentation on large-scale computational models (e.g., CAD/CAM models involving thousands of parameters; see Lindsay, Kettenring, and Siegmund 2003; 2004). • New clustering and classification methods for data-mining applications (see Hastie, Tibshirani, and Friedman 2003). 611 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. In addition, the increasing availability of user-friendly statistical software and readily accessible information about statistics continues to make statistical methods increasingly accessible to nonstatisticians. This will further challenge statisticians to develop tools that are as simple as possible to use – and maximally robust to potential misuse. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.7.2 Advances in Application Areas Bioinformatics and Statistical Genetics The rapidly developing field of bioinformatics has been loosely referred to as any use of computers to handle biological information (see Tramontano 2006). Gene mapping, a key application area, has “the ultimate goal... to identify the genes that play important roles in the inheritance of particular traits and to explain the role of these genes in relation to one another and... the environment” (see Siegmund and Yakir 2007). This typically requires the assembly and statistical analysis of large sets of DNA and other information. Genetic research experiments, for example, now routinely collect over 300,000 measurements per subject. Nanotechnology The ability to manipulate material characteristics at the individual molecule level is leading to important changes in the way products are designed, manufactured, and used. The routine clinical use of nanotechnology to deliver drugs to different locations within the body is an active research area. The need to build high-quality and high-reliability products on a nanoscale creates statistical challenges that seem to resemble those faced, early on, by the semiconductor industry in raising yield. (See Dasgupta, Ma, Joseph, 612 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Wang, and Wu 2007; Nembhard 2007; Rue 2006 for further discussion.) Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. Automated Tracking Implanted health-monitoring devices to provide early warning of, say, an impending heart attack hold much promise. As in other applications of remote monitoring, a major contribution of statistics is to help establish action thresholds that provide the desired balance between maximizing detection sensitivity and minimizing the false alarm rate. Other Areas People all over the world have come to depend on increasingly complex systems (e.g., economy, health, and transportation) that utilize statistics. For example, dynamic travel routing might be extended from using information on current congestion to using predictions of future conditions. These might employ a statistical algorithm that uses data on past traffic (by time of day and day of week), the impact of (forecasted) weather and of special local events (e.g., a football game) to predict traffic volume. Statistical methods also help address such “hot topics” as climate change and the assessment of world energy reserves. In addition, the continuing flow of government and industry regulations calls for new programs of disciplined data gathering and statistical analysis. The Sarbanes–Oxley Act, a U.S. federal law enacted in response to various corporate and accounting scandals, provides an example. Thus, Faltin and Faltin (2003) illustrate how a Six Sigma approach, employing publicly available data and statistical 613 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. tools, could have predicted WorldCom’ s ultimate bankruptcy. Statistics will, undoubtedly, continue to make important contributions to addressing new technical and societal challenges involving business and industry as these emerge; in turn, new applications will stimulate further powerful developments in statistics. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 13.8 MAJOR TAKEAWAYS • Applications of statistics in the food, beverage, and related industries are characterized by: • The need (especially for processed foods) to minimize the variability that customers experience from one purchase to the next. • Data for product assessment are obtained from consumers, instruments and experts and are often ordinal or categorical. • The applicability of mixture designs in planned experiments. • Manufacturing issues similar to those encountered for many continuous processes. • High concern with food safety and spoilage. • The semiconductor business is characterized by its many process steps and efforts to improve yield. It is heavily data-intensive and provides numerous opportunities for the use of statistics. Applications include: • Studies to understand the impact of process variables upon performance • Monitoring and controlling critical process variables • Product life assessment • Product burn-in • The communications industry, another highly data-intensive business, has traditionally been a heavy user of statistical 614 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. methods. Typical problems involve usage assessments, reliability and service improvement, and business assessments. The emergence of the Internet has led to new applications. • Statistical methods play a prominent role in image analysis for medical, security and other applications including: • Restoration of images, such as those obtained in digital photography, to remove defects. • Pattern recognition in, for example, diagnosing diseases by medical scanning or identifying criminals by matching their pictures with those in an available database. • Statistics also makes major contributions in many other application areas that often involve business and industry, including health care, epidemiological studies, national security, space exploration, oil exploration, environmental studies and chemometrics (the analysis and modeling of chemical data). • Technological advances in both information sciences and application areas continue to lead to new applications in bioinformatics and statistical genetics, nanotechnology, automated tracking, and numerous other areas, and to provide the impetus for the development of powerful new statistical methods. DISCUSSION QUESTIONS General 1. Why do the food, beverage and related industries “provide an ultimate example of the need for robust design,” as stated in Section 13.2.2? 2. Assume that you are the manufacturer of a popular brand of chocolate chip cookies. Propose a program to help ensure product consistency. 615 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. 3. Assume that you are the manufacturer of a canned fish product. Propose a program to help ensure product safety. 4. How does the “relatively long elapsed time to conduct even a single production run” (Sidebar 13.3) in semiconductor manufacture impact the statistical evaluations? 5. Practices in use in the semiconductor industry, such as end-of-line inspection, accepting relatively low product yield, and burn-in, tend to be contrary to the goals of ensuring up front high quality and reliability. Discuss this apparent disconnect and why and when it might be appropriate for the semiconductor industry. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 6. A telephone company has asked you to develop a usage assessment system to provide information to improve service. What would you propose? 7. Suggest some applications of statistical image analysis beyond those presented in this chapter. 8. What do you think are the most important contributors to uncertainty in a face recognition system? 9. Suggest some measurable characteristics associated with the correct administration of prescription drugs in a hospital. 10. Applications such as pattern recognition (Section 13.5.3) and oil exploration (Section 13.6.5) call for developing a scheme that strikes the best possible balance between incorrectly identifying a signal when it is not 616 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. present (Type I error) versus failing to identify a signal when one exists (Type II error). Explain this comment and suggest what might be done to increase the chances of making the right call. 11. Suggest and briefly describe some likely future applications of statistics in business and industry beyond those discussed in this chapter. Technical Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 1. We state in Section 13.2.2 that to achieve homogeneity in processing roasted peanuts, “measurements on such factors as sugar content and moisture are obtained on crop samples during harvesting. This information is then used to set processing (e.g., roasting) conditions.” Explain this concept and propose a plan for implementation. 2. In the discussion of the need for reducing the variability in the processing of flour in Section 13.2.2, we state that “Homogeneity in the final product is attained by adjusting the mixing proportions, based upon product or process measurements.” Propose a system for doing this. 3. Propose a GRR study to help ensure consistency in judgment over time and between tasters for end-of-line wine tasting. 4. Obtain further details about the HACCP system for proactively ensuring food safety (Section 13.2.6) and discuss how statistics fits in. 617 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. 5. In Section 13.3.6, we state that “Statistical concepts help determine how long units should be burned in, based upon cost considerations.” How would you do this? 6. Variance components analysis was suggested in Sidebar 13.4 “to quantify the relative contribution to total variability of, say, lots, wafers, positions within a wafer, individual units and measurement error.” This suggests that each of these are random, rather than fixed, effect (analysis of variance) variables. Discuss this assumption, especially with regard to wafer position, and how the analysis is changed if the assumption is not satisfied. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 7. Consider a Bayesian interpretation of the face recognition example (Section 13.5.3). Given a defined set of N prior probabilities and N likelihood values from a database of N images, define the posterior probability of a true match. How can you adjust the prior probabilities to change system sensitivity and specificity? 8. What kind of statistical models might you use to model different contributors to uncertainty in a face recognition system? 9. Pattern recognition (Section 13.5.3) and chemometrics (Section 13.6.7) both address problems that involve comparison of an unknown sample against known ones to find a match. Discuss the similarities and differences in the two problems and the statistical methods used to address them. 10. We state in a footnote to Section 13.6.6 that in modeling pollution, “a lognormal distribution is often used 618 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. to represent discharge amounts.” Why might a lognormal distribution be appropriate in this situation? 11. Select one or more of the “Other Applications” described in Section 13.6. Describe statistical evaluations that might likely be performed. 1. This discussion has benefited greatly from inputs provided by Lynne Hare and Mark Vandeven. 2. In contrast, expert evaluators are generally selected based on their proficiencies. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 3. Due to fatigue, etc. it is sometimes inadvisable to have each evaluator assess each product. This leads to the use of randomized incomplete block experimental designs to obtain data for making comparisons. 4. T he U.S. media recently reported an E. coli outbreak, attributed to raw spinach, and peanut butter contaminated with salmonella. 5. Per the HACCP website http://www.cfsan.fda.gov/ ~comm/haccpov.html. 6. This discussion has benefited greatly from inputs provided by Mike Clayton, Veronica Czitrom, Mary Lanzerotti, Paul Tobias, and Emmanuel Yashchin. Various definitions were adapted from the Semiconductor Industry Association’s online glossary. 619 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46. 7. A more advanced approach involves monitoring individual estimated variance components (see Yashchin 1994). 8. This discussion has benefited greatly from inputs provided by Diane Lambert and Scott Vander Wiel. 9. References to articles that discuss specific application areas, beyond those cited here, are to be posted on the book’s ftp site. 10. This section was written by Colin McCulloch. 11. The term “Gaussian distribution” is widely used in the imaging literature for the normal distribution. Copyright © 2008. John Wiley & Sons, Incorporated. All rights reserved. 12. A lognormal distribution is often used to represent discharge amounts. 620 Hahn, G. J., & Doganaksoy, N. (2008). The role of statistics in business and industry. ProQuest Ebook Central http://ebo Created from usf on 2021-05-19 19:26:46.
Purchase answer to see full attachment