Top-down approach example

Top-Down/Bottom-Up Approach

Once the system and goals have been defined, the analysts and the management team need to decide if they are to develop a top-down or bottom-up approach to understanding the system whose reliability is to be improved.

Selecting Health Items or Attributes

A top-down approach has been used when developing the majority of the existing health instruments. That is, their content has been derived from the existing body of literature, instruments, and health surveys. Patient or public involvement, if any, occurred in the later stages, when an instrument was being validated. In general, there is a lack of awareness that careful selection of key health items [or attributes] is paramount in developing a sound health measurement instrument. That said, a nice example of an exception to this rule is the CHU9D [Chen etal., 2015; Stevens, 2016].

Developers of health-outcome instruments have realized that content validity cannot be established through quantitative psychometric analyses. This point was highlighted in the FDA PRO guidance [2009]. Face and content validity are important aspects of health-state instrument development, yet they have seldom been appraised. An appropriate selection of health items can be enhanced by acknowledging the importance of clinical insight and patient participation to the development process. Since the phenomena to be measured are often complex and intangible, it is likely that salient health aspects will be missed if scientists with little or no experience of illness and disease are in charge of deciding which items are important and should be incorporated in an instrument.

At an early stage, patient input is of utmost importance to identify the concepts and craft the wording for the items of a health instrument. After drafting the instrument, the [often required] second step is to conduct cognitive interviews. At that point, the patients fill in the instrument and then comment on its relevance, clarity, and comprehensiveness. This qualitative part of the research should be conducted in samples of patients from the target population to establish content validity for a particular use. The rationale is that while an instrument may have content validity for one disease, it might omit concepts that are important when assessing another disease [Patrick etal., 2011a,b; Stevens and Palfreyman, 2012].

Until now, most content for health-status instruments has been generated in consultation with expert panels or by a synthesis of existing instruments. Probably a more valid and certainly a more patient-centered approach would be to organize focus groups. All aspects of health that are important to the patients may be uncovered in these sessions until saturation has been reached. It is reached at the point when no additional aspects are raised in a subsequent focus group. Another approach is to present a large patient target group with a complete overview of relevant health attributes that have been selected after theoretical reflection and in-depth study of existing health-outcomes instruments, followed by sessions with focus groups, when the participants are asked to select the most important attributes based on their health experience.

Such a systematic approach is different from quantitative methods. In fact, most of this book deals with quantitative research. However, many quantitative methods are inappropriate for qualitative issues. Sometimes researchers apply statistical techniques to determine which items are important. For example, factor analysis is often used for this purpose. However, factor analysis is directed by relationships [i.e., correlations] between variables [i.e., items]. This means that two items with more or less equal distributions of responses [i.e., frequencies] on the response categories will load on the same factor. Therefore, the results of factor analysis will not tell us anything about the importance of these items. To find out what the important items are, we have to conduct qualitative research or studies with specific response tasks to generate data that can be properly analyzed. Response data collected on the basis of patient exercises can be analyzed by special procedures to highlight specific relationships and clusters. One such procedure is Q methodology [Stenner etal., 2003].

Since the phenomena are often complex and intangible, important aspects are likely to be missed by the blind application of the statistical or psychometric approach. An essential ingredient is lacking in the psychometric approach [and certainly in the preference-based approach]: clinical and patient insight, which even sophisticated statistics cannot provide. Neither a quantitative nor a qualitative approach alone seems sufficient [Zyzanski and Perloff, 1999; Fava etal., 2004].

Prototyping

The publication of the prototype using a drill-down procedure will be the first time that the future users will actually see the product, so it is important that both the contents and the appearance are first class. It should not be treated as just a draft. In the words of the proverb, You dont get a second chance to make a first impression.

Another reason for creating some early detailed procedures is in response to incident in particular areas or on particular types of operation.

Guidance within Employee Development Schemes

A contrasting top-down approach to workplace guidance initiated by employers was found in terms of the guidance elements of an employee development scheme in a Scottish university, known as Learning Works. Under the scheme, manual, technical, ancillary, clerical, and secretarial staff all placed on the lower points of the Universitys pay scale may apply for an annual learning allowance of £150. A system of site coordinators across 25 different locations in the university exists, and these volunteers, usually previous beneficiaries of the scheme, encourage fellow workers to consider taking up its benefits by offering some initial advice and information. Thus, on a peer basis, they provide a signposting role similar to that of the learning representatives in the UK or the guidance corners in Denmark. In addition, Learning Works employs a part-time learning adviser, who is a qualified career guidance officer with experience of working with adults, to provide impartial guidance to anyone considering taking up a learning allowance [Plant and Turner, 2005].

4.09.1.2 The Roots of Top-Down Analyses of Associative Memory in Lymnaea

Shortly after the publication of the first papers describing CPG interneurons, motor neurons, and modulatory neurons of the feeding network [Benjamin and Rose, 1979; Rose and Benjamin, 1979; McCrohan and Benjamin, 1980a,b], seminal behavioral work established the ability of Lymnaea to form long-term associative memory after a few pairings or even just a single pairing of a nonfood chemical stimulus and a food stimulus [Alexander et al., 1982, 1984; Audesirk et al., 1982]. It was subsequently shown that Lymnaea is also capable of forming a positive association between tactile stimuli and food, but only after multiple trials [Kemenes and Benjamin, 1989a]. Both the chemical and tactile conditioning paradigms revealed a number of characteristics of associative learning in Lymnaea that were shared by learning in vertebrates, such as dependence on age and motivational states, stimulus generalization, discriminative learning, and classical-operant interactions [Audesirk et al., 1982; Kemenes and Benjamin, 1989a,b, 1994]. Later on it was also demonstrated that Lymnaea can form negative associations between a food conditioned stimulus [CS] and an aversive chemical unconditioned stimulus [US] [classical conditioning] [Kojima et al., 1996] or between a behavior and an aversive tactile stimulus [operant conditioning] [Lukowiak et al., 1996]. The wealth of knowledge about the behavioral features of both operant and classical, aversive and reward conditioning, together with a detailed understanding of the neuronal networks underlying the unconditioned behaviors used in these paradigms [feeding and respiration, Figure 2], has made Lymnaea a very attractive experimental model for top-down analyses of learning and memory.

The two fields of research, physiological analysis of CPGs and behavioral analysis of associative conditioning, started to converge in the mid- to late 1990s when neuronal correlates of both classical and operant conditioning were first described [Whelan and McCrohan, 1996; Kemenes et al., 1997; Kojima et al., 1997; Staras et al., 1998, 1999; Spencer et al., 1999]. The most important neuronal aspects of associative learning in Lymnaea are discussed in another chapter [See Chapter 1.30] as well as several review articles [Kemenes, 1999; Lukowiak and Syed, 1999; Benjamin et al., 2000; Lukowiak et al., 2003b]. Detailed analyses of molecular mechanisms of associative memory started in the late 1990s after sufficient information had been obtained on both the behavioral and neuronal aspects of associative learning [Figures 1 and 2].

Questions raised by the top-down approach

These are general questions that relate not only to CIs but also to other neural prostheses. Tentative answers in the context of CIs have been presented in this chapter. Each of the questions can of course be restated as a hypothesis or a set of connected hypotheses. Studies are now underway to evaluate some of the various hypotheses, in tests with CI users and control subjects. In addition, questions 5 and 6 are important in contemporary neuroscience, so those questions are receiving considerable attention from a large number of research teams worldwide. Further information in response to all of the questions listed above may well help to shepherd in a new era in neural prosthesis designs, in which the brain is regarded as a key part of the system.

2.4.1 Logic and Virtual Machines

One example of a top-down approach is the view that logic is adequate to describe psychological phenomena. This view is attractive: it provides epistemological foundations [the axioms of the logic]. However, as we shall see [section 4.1], automatic reasoning based on logic is not enough for a general psychological account. We therefore need different logics for different contexts.

Nevertheless, the idea continues to influence. Although well-defined formalisms are not thought to be good at describing neural computation, there is a view that at some level of abstraction logic is the best description of inference systems. There is therefore the view that it should be possible to describe any such system as a logical calculator.

Different levels of description of computational systems are termed virtual machines. A key question about neuroscience and artificial neural nets, therefore, is whether such substrates can provide material for a virtual machine that supports logic-based reasoning and subjectpredicate descriptions of the environment.

Unannounced Audits

Wherever a company may be with respect to the status of its risk management programs, it is important to be ready for an unannouced audit. There is no telling when an accident might happen, and, if the accident is bad enough, there will be audits from regulators and attorneys representing various plaintiffs. The fact that a facility is in the middle of developing its risk management program is not a valid excuse; an accident can happen at any time, therefore, an audit can take place any time.

The practical implications of this understanding are twofold. First, it means that all information to do with risk management must be properly organized and indexed so that it can be quickly located. Doing so will create a good impression with the auditor, and it will reduce the possibility of not being able to find the information that was asked for.

A second implication of being ready to meet an audit at all times is that it strongly encourages a top-down approach to developing a risk management program. Initially, there will be very little detail in the program. But, as it develops more detail will be added. The key is that, at all times, the program will be complete, and it will not be a collection of miscellaneous documents that may or may not be connected to one another in a coherent manner.

Theory-driven, top-down approach

A theory-driven, top-down approach capitalizes on identifying features a priori, reducing the costs to researchers by making the task of identifying if then profiles easier and more efficient. Indeed, such approaches typically allow researchers to skip steps 2 and 3 [and often step 4] in Fig. 3. One example of a top-down HRWP approach is a study on dating preferences contributing to psychologically abusive relationship dynamics [Zayas & Shoda, 2007]. The researchers used theory to identify the features of dating partners that may attract women who might be at risk for experiencing psychological abuse. The researchers identified a priori characteristics such as partner jealousy, controlling, and withdrawing behaviors as features of a potential dating partner and ensured that the dating profiles presented as stimuli varied systematically on the presence of these features. The researchers found that women who had a history of experiencing psychological abuse were twice more likely to select a potential dating partner with red flags, i.e., possessing personality traits associated with an abusive partner. In a second study, Zayas and Shoda [2007] used a similar approach to examine the features of dating partners that attract men who may perpetuate psychological abuse. They found that men who had a history of perpetuating psychological abuse in their past relationships were three times more likely to select a potential female dating partner characterized by attachment insecurity [i.e., high vulnerability and low self-esteem].

5.2.1 The entities

FRBR models the bibliographic data using entities to represent data that are distinctly identified. It also uses relationships to establish the associations between the different entities. Examples of entities are person and work. These entities are related, as one or more persons are considered responsible for one or more works. To build an ER model the designer has to identify the entities, the characteristics of the entities [e.g. work has a title and person has a name] and the relationships between the entities.