Flashbulb Memories

New Challenges and Future Perspectives

Olivier Luminet, Antonietta Curci

• Publisher: Psychology Press
• Year: 2017
• ISBN:
  9781317226147
• Edition: 2
• Topic: Psychology
• Subtopic: History & Theory in Psychology
• Index: Psychology

Are Flashbulb memories special or ordinary memory formations? Are emotional, cognitive, or social factors highly relevant for the formation of Flashbulb memories? How can sociological, historical, and cultural issues help us to understand the process? What is the difference between Flashbulb memories, memories of traumatic experiences, and highly vivid personal memories? How can we provide a valid and reliable measure for Flashbulb memories?

This edition of Flashbulb Memories: New Challenges and Future Perspectives revisits these questions, considering significant new evidence and research in the field. It now includes additional chapters focusing on experimental investigations, and review studies on positive vs. negative Flashbulb memories.

Bringing together leading international researchers, the book presents significant progress in this area of research, which has remained divisive for the past 40 years. The discussion of Flashbulb memories also contributes to the understanding of the general functioning of autobiographical memory. It will provide essential reading for researchers in Flashbulb memories and will be of great interest to those in related areas such as cognitive psychology, social psychology, cross-cultural psychology, sociology, political sciences, and history, as well as clinicians dealing with those who have strong Flashbulb memories after personal traumatic events.

1
Measurement issues in the study of Flashbulb memory

Antonietta Curci

Measurement models have carried significant assumptions – either explicit or implicit – about the nature and inclusion of FBMs in the domain of autobiographical memory. The chapter compares the dimensional (Latent Trait Analysis) and categorical (Latent Class Analysis) measurement models of FBM and shows that dimensional models fit well with ordinary autobiographical data, while FBMs are better modelled through a categorical approach. The adoption of a taxonomic model to a large dataset on the September 11th attacks provides empirical support to the idea that FBMs represent a discontinuity in the field of autobiographical memory, and lead to important conclusions on their special nature.

Measuring Flashbulb memory (FBM) is one of the key issues in the investigation of the phenomenon. Put in another way, it is the problem of construct validity (Carmines & Zeller, 1979) applied to the investigation of FBM. Methodologists usually recommend a careful consideration of the degree to which a given measurement model matches that which it claims, or purports, to be measuring. In doing so, a clear theoretical definition of the construct under analysis needs to be provided along with an accurate specification of the employed measurement model. In the field of FBM, researchers have adopted different theoretical views regarding the phenomenon (see also Wright & Carlucci, Chapters 2; Talarico & Rubin, Chapter 4, this volume). Both explicitly and implicitly, measurement models of FBM have carried significant assumptions about the nature and inclusion of the phenomenon in the general domain of autobiographical memory. The present chapter firstly provides an overview of different measurement models traditionally adopted in assessing FBMs, by showing the limitations of these models and the various attempts to overcome these. Subsequently, a brief outline of the peculiarities of both dimensional and categorical models will allow the reader to evaluate their appropriateness in the investigation of FBM. Finally, from both the theoretical and empirical point of view, similarities and differences between FBMs and other forms of ordinary autobiographical memories (such as memory for the original event) are considered. Generally, categorical models appear to best account for FBMs as particularly vivid and detailed autobiographical memory formations. A practical implication for research is the need to improve FBM assessment to take into account the categorical nature of the construct, also relying upon the current availability of many advanced statistical tools.

The measurement of FBM

The problem of measuring FBMs dates back to the first research work on the topic. In their original paper, Brown and Kulik (1977) defined FBMs as memories from the attributes of the reception context of shocking public news. In other words, people may retain, for a long time, not only the original event itself, but also the reception context for this event, that is, the place where they were, the time when they learned of the event, their ongoing activity, the informant, the personal reactions and reactions of others, and the aftermath of the event (Bohannon, 1988; Brown & Kulik, 1977; Conway et al., 1994; Larsen, 1992). The present chapter reviews studies focusing on the way manifest indicators of the reception context of a shocking event are related to the latent construct of FBM. Indeed, when assessing FBMs, researchers not only make assumptions concerning the specific features to be measured (vividness, consistency, confidence, longevity, etc.), but also on the mathematical model connecting these observed features with the latent construct. This, in turn, has noteworthy implications for the theoretical advancement of research on FBMs.

In their original study, Brown and Kulik (1977) operationalized FBMs in two ways. First, individuals were scored as having an FBM if they answered “yes” to the direct question: “Do you recall the circumstances in which you first heard that…?”. Second, an FBM was identified if people could remember at least one attribute of the reception context (Brown & Kulik, 1977; Pillemer, 1984). In assessing the phenomenon, Brown and Kulik (1977) assumed that a simple counting of attributes of the reception context would represent a good approximation to the construct of FBM. When applying such a procedure, the outcome of the assessment is an absolute scale, in principle without upper boundary, since in real life the total number of members of a collection cannot be a priori defined (Luce & Suppes, 2002). Thus, the number and characteristics of attributes chosen to define FBM is completely arbitrary. Brown and Kulik (1977) selected six attributes of the context, which they considered more indicative of the nature of FBMs, and all six attributes were given the same relevance in their model. However, the authors’ choice was neither theoretically justified nor empirically supported. Unfortunately, a simple counting of attributes is completely uninformative of the intrinsic nature of the construct in analysis. In spite of this, the procedure of summing up FBM attributes (Brown & Kulik, 1977) was subsequently adopted by researchers (Bohannon, 1988; Kaya Kizilöz & Tekcan, 2013; Kvavilashvili et al., 2003; Pillemer, 1984; Talarico & Rubin, 2007).

Other studies have attempted to reduce the major flaw of Brown and Kulik’s measurement model (1977) by prioritizing specific details of the context. Winograd and Killinger (1983) identified an FBM from the mention of ongoing activity in participants’ recollections. Wright (1993) required that each participant recalled at least one attribute from location, other present people, and ongoing activity. Despite having evident limitations, these approaches introduced the idea that FBM is more than a simple collection of irrelevant attributes, since some of them are more representative of the construct than others.

Neisser and Harsch (1992), in their study on the Challenger disaster, employed a procedure called WAS (Weighted Attributes Scores), which assigned different weights to different attributes of the reception context. The authors assumed that the attributes were not all equally important. Some of them were defined as “major” (i.e. location, informant, and ongoing activity), since they seemed to be essential to identifying the reception context. Other attributes were considered “minor” (i.e. other present people, and time), since one could be wrong on these and still essentially accurate about the major details. The WAS procedure consists of assigning a score for each recalled major attribute, plus a bonus point when the subject scores above a given threshold on the set of minor attributes. More specifically, scores ranging from 0 to 2 were assigned to the individual’s memories for location, informant, and ongoing activity. Furthermore, an additional score of 1 was added if the participant scored at least 3 on the indicators assessing other present people and time (on two scales also ranging from 0 to 2). As a consequence, for each individual, the final WAS measure for FBMs ranged from 0 to 7 (Neisser & Harsch, 1992). The WAS system was subsequently used in more recent studies on memory for the September 11th attacks (Kvavilashvili et al., 2010; Kvavilashvili et al., 2009; Pezdek, 2003; Shapiro, 2006; Smith, Bibi, & Sheard, 2003; Tekcan et al., 2003), and it represented a clear advancement towards a measurement model which considered FBM as a qualitatively different phenomenon from ordinary memory formations.

More recently, Curci et al. (2001), Curci and Luminet (2006), and Luminet et al. (2004) modelled FBM data through a statistical technique called CatPCA (Categorical Principal Component Analysis; see Gifi, 1990), which shares the same logic of the WAS approach (Luminet et al., 2004). CatPCA is a principal component analysis specifically aimed at scaling categorical or ordered categorical variables (van de Geer, 1993). In employing this technique, the authors aimed at getting a composite measure, which combines scores of different indicator variables corresponding to the attributes of the reception context, weighted with respect to their relevance in identifying FBMs. With respect to WAS, the main advantage of CatPCA is that the weights assigned to the scores are not decided a priori by the researcher, but derive from the empirical distributions of the indicator variables in the sample of respondents in the study (Greenacre, 1993). Luminet et al. (2004) discussed the similarities and differences between the WAS and CatPCA approaches, showing that, although the scores for the two procedures were highly correlated, CatPCA was preferred because its scores are obtained from an empirical basis and not from researchers’ aprioristic assumptions.

In sum, the literature reviewed thus far substantiates some relevant points: (1) Traditional measurement models appear inadequate to assess FBMs; (2) Evidence concerning the measurement of FBM have implications for the theoretical understanding of the nature of the phenomenon; (3) Significant implications can be derived for a general model of autobiographical memory.

The assessment of latent constructs

Dimensional vs. categorical models

Researchers in psychology usually deal with unobservable, or latent, constructs. For instance, intelligence, introversion, and mathematical ability cannot be directly measured as can height or weight, since they are concepts rather than physical entities. Nevertheless, some empirical attributes can be considered as observable indicators of underlying latent constructs (Baker, 1985). One basic assumption of latent variable models is the local independence of the indicator variables given the latent variable(s) (Lazarsfeld & Henry, 1968). Postulating the local independence means that, after removing the variability between the latent construct and its observed indicators, the individual’s responses to the items of a test are statistically independent of each other (Hambleton, Swaminathan, & Rogers, 1991; McCutcheon, 1987).

The so-called dimensional models assume that a continuous, normally distributed latent construct accounts for variations among observed indicators (Moustaki, 1996). For instance, the ability to provide correct answers to a set of mathematical calculations might be conceptualized as a continuous trait (i.e. the mathematical ability) which each individual holds to a different degree. Psychologists are familiar with factor analysis models, which consider observed variables as linear combinations of latent factors, plus random error terms. Factor analysis models are most useful with continuous indicator variables, for which correlation coefficients can easily be computed. Fitting these models involves finding the values of the latent variable parameters which maximize the probability of reproducing a correlation matrix of indicators as close as possible to the observed product moment correlation matrix (Comrey & Lee, 1992; Kim & Mueller, 1978).

When the aim of the researcher is to reduce a set of categorical observed indicators (binary, ordered categorical, simply categorical) into a smaller set of latent factors, the Latent Trait Analysis (LTA) is one of the most accepted dimensional approaches (Bartholomew et al., 2002; Rost & Langeheine, 1997). In the area of educational testing and psychological assessment, LTA is termed Item Response Theory (IRT). A considerable overlap exists between LTA and IRT, leading the authors to consider them as basically interchangeable approaches. Among psychologists, the acronym IRT is more popular, and it will be preferred in the following pages.

Item Response Theory has two basic postulates: (1) The individual’s performance on a test item depends on some latent abilities called traits; (2) The relationship between the individual’s item performance and latent traits can be described by a monotonically increasing function (Embretson & Reise, 2000; Hambleton et al., 1991). This function, called an IRT Item Characteristic Function, expresses the probability of an individual giving a correct answer to an item of a test as a function of his/her ability (trait) (Baker, 1985). IRT methods were originally unidimensional, assuming that only one latent trait accounts for variations of observed indicators, but generalizations to multidimensional models have been more recently proposed (Embretson & Reise, 2000). In the present chapter, unidimensional IRT models will be discussed, since the application of IRT methods to FBM data has been limited to this approach (Curci, 2005; Curci & Lanciano, 2009; Wright, Gaskell, & O’Muircheartaigh, 1998). The theoretical rationale of the models and some details on the procedure will only be presented here, since a technical discussion of their mathematical features is beyond the scope of the present chapter (for more details, see Bartholomew et al., 2002; Embretson & Reise, 2000; Hambleton et al., 1991).

Unidimensional IRT models differ from each other with respect to the form of the Item Characteristic Function, and the number of parameters in the model. For example, in the investigation of a student’s mathematical ability, a starting assumption is that, in providing the correct answers to a set of mathematical computations, each individual possesses some amount of the ability under investigation. In other words, each student has a hypothetical score on the scale of a latent construct corresponding to her/his mathematical ability. This construct cannot be directly assessed, but, for each student, its amount is inferred from the number of correct responses provided to the set of mathematical computations. IRT models allow the researcher to estimate the probability of an individual to choose some response categories from a set of observed indicators (i.e. correct responses to the mathematical computations) as a function of an underlying latent trait (i.e. the mathematical ability). In doing so, IRT models describe each observed indicator with respect to some properties (difficulty, discrimination, etc.) which define the position of the individual along the continuum represented by the latent trait.

If we denote the ability score as theta, then, at each level of theta, a different probability is associated to obtaining the correct answer to a given computation. This probability is denoted as P(theta). For each item of the set, plotting P(theta) as a function of theta will result in a smooth S-shaped curve, called an Item Characteristic Curve (ICC). At the lowest levels of theta, the probability of a student providing the correct answers to the considered mathematical computation is very low, and it increases as the amount of possessed mathematical ability increases following the shape of the ICC. The one-parameter logistic model (or Rasch model, Rasch, 1960) simply considers the location of the curve along the x-axis, by including in the equation for P(theta) only the so-called difficulty parameter. The higher the difficulty parameter for a given item, the higher the ability required for correctly answering that item, the more the ICC will be located along the right side of the x-axis. In Figure 1.1, the ICC on the left has a lower difficulty parameter than the ICC on the right, and this means that less ability is required in order to provide the correct answer to the item represented by the former ICC as compared with the item represented by the latter. To illustrate, much more mathematical ability is required to solve an algebraic equation ...

Table of contents

• Cover
• Title
• Copyright
• Contents
• Figures
• Tables
• Contributors
• Acknowledgments
• Introduction: How research on Flashbulb memories has developed in the last ten years
• 1 Measurement issues in the study of Flashbulb memory
• 2 Flashbulb memory methods
• 3 Using Structural Equation Modelling approaches to better understand the formation of Flashbulb memories
• 4 Ordinary memory processes shape Flashbulb memories of extraordinary events: A review of 40 years of research
• 5 The consequences of consequentiality
• 6 When a flash is caught in a lab: An experimental approach to the investigation of Flashbulb memories
• 7 Flashbulb, personal, and event memories in clinical populations
• 8 A comparison of Flashbulb memories for positive and negative events and their biopsychosocial functions
• 9 Flashbulb memories and social identity
• 10 Aligning Flashbulb and collective memories
• 11 Flashbulb memories and collective memories: Psychosocial processes related to rituals, emotions, and memories
• 12 Culture in Flashbulb memory
• 13 The study of Flashbulb memories: The good, the bad, and the way forward
• Index