• INTRODUCTION
• THE MODEL
•   Coherence optimization and constraint satisfaction
•   Cognitive unit sets
•   The relevant CUSs and the relations between them
•   Process
•   Social level
• SIMULATIONS
•   Suicidal rates: constancy and variability
•   Group differences
• DISCUSSION
• REFERENCES

Abstract

This article describes a model that approaches both individual and social dimensions of suicide. At the individual level, the model construes suicidal thoughts as optimization of coherence between relevant cognitive unit sets (i.e. collections of interrelated cognitive units that focus on a particular external or internal object). It assumes that suicidal thoughts are mostly affected by the manner individuals construe their egos, their economic, relational, professional, and health situations, as well as life, death, and suicide.  At the social level, the model argues that suicide rates depend on the distributions of these cognitive unit sets and of the connections between them. The model has been implemented in a computer program that simulates important characteristics of suicide rates.
 
 

INTRODUCTION

In Dostoyevsky's novel, The Possessed, one of the characters, Aleksei Kirillov, maintains that humans have an absolutely free will and that only a gratuitous suicide, which transcends all psychological motivations, would be able to express it. In spite of its philosophical attractiveness, extremely few scientists would favor the idea of a gratuitous suicide act and would rather prefer to search for suicide’s causes and mechanisms. Many of them would say that suicide is determined by individual factors. Psychologists, whether they construe suicide as escape from psychological conflict or focus on the beliefs, expectations, problem solving skills, developmental factors or other psychological variables that affect suicide, usually adhere to this conception. Durkheim, the author of the most influential theory of suicide (Atkinson, 1978), did not. According to his counterintuitive theory, although suicide seems the most personal act a human being can do, it is exclusively determined by social factors (Durkheim, 1897/1993).

The split between approaches centered on individual factors and approaches centered on external factors (mostly on social factors) is one of the most important in suicidology. Usually, researches that focus on the individual dimension of suicidal behavior are not concerned with and are of no use to researches that focus on the social dimension of suicide and vice versa. However, there are researches concerned with both individual and social aspects of suicide. The classic theory elaborated by Henri and Short (1954) provides a nice example. Its basic psychological assumptions are simple: frustration is positively related to aggression and people attribute the responsibility for their frustration externally or internally, external attributions determining aggressive acts oriented toward others (homicide) and internal attributions determining self aggressive acts (suicide). A negative economical evolution of the society increases frustration among individuals and, consequently, the number of aggressive acts. However, suicide and homicide rates do not rise uniformly. Suicide will be more frequent among those with high status and homicide will be more frequent among those with low status because high status is associated with weak external restraints that determine  internal attributions of responsibility, and low status is associated with strong external restraints that determine external attributions of responsibility.

Like Henri and Short’s theory, the present work addresses both psychological and social dimensions of suicide. Unlike Henri and Short’s theory, the theoretical frame used here is different. At the individual level, the model focuses on the optimization of coherence among relevant cognitive unit sets. At the social level, it assumes that suicidal rates depend on the relevant cognitive unit set distributions of a population.
 
 

THE MODEL
 

Coherence optimization and constraint satisfaction

In a coherence optimization problem, one has to divide a set of elements that constrain one another into an accepted set and a rejected set, in a way that maximizes the satisfaction of constraints. The satisfaction is not uniform, constraints that are more important being preferred. A constraint between two elements is positive if they fit together (i.e. the relation between them can be of positive association, consistency, facilitation, explanation etc.) and negative if they conflict (i.e. the relation between them can be of negative association, inconsistency, incompatibility). If the constraint between two elements is positive, they are accepted or rejected together, if the constraint is negative one is accepted and the other rejected (Thagard and Verbeurgt, 1998; Thagard and Kunda, 1998).

Constraint satisfaction networks are used in order to approximate coherence optimization. They have been successfully used in psychology to model various cognitive (Thagard and Millgram, 1995; Holyoak and Thagard, 1997 etc.) and social psychological processes (Shultz and Lepper, 1996; Kunda and Thagard, 1996 etc.). In a parallel constraint satisfaction network, the basic unit is an oversimplified analogue of the biological neuron. It is defined by three characteristics: input set (the activation coming from other units, weighted with the connection importance), activation function (how is processed the input set), and activation (the result obtained after the activation function has been applied to the input set). The connection between two units represents a constraint. Negative connection weights correspond to negative constraints and positive connection weights correspond to positive constraints. Constraints that are more important have greater connection weights, while less important constraints have smaller connection weights. Goodness stands for the degree to which constraints between units are satisfied. The goodness contributed by a particular unit i to the overall goodness is given by:

where, w_ij is the weight of the connection between units i and j, a_i is the activation of the unit i and a_j is the activation of unit j. The overall goodness is the sum of individual goodnesses:
 

Maximizing this goodness function is equivalent to optimizing coherence. The network goes through repeated cycles in which units are randomly selected and updated according to an activation function. There are several equations that constraint the possible range of unit activation between some limits. The connectionist program that implements the coherence model of suicide uses the following equation:
 

where, net_i represents the net input to unit i and is given by:
 

This activation rule, also known as the logistic or sigmoidal function is usually used as a probabilistic function (Dumitrescu and Costin, 1995) but, recently, it has also been used deterministically (Todereanu, Costeiu and Giurgiu, 1995). The present model uses the logistic function in this later manner because it simplifies the calculus.

After several cycles of unit activation adjustment, the network settles down, that is, units will no longer significantly change their activation values. If the final activation of a unit exceeds a specified threshold, the unit is accepted. If not, the unit is rejected. Here, units are encouraged to reach maximum or minimum activation values: units with maximum final activation are accepted, whereas those with minimum final activation are rejected.
 

Cognitive unit sets

Some psychological and psychiatric studies of suicide focus on variables such as personality or temperament traits (e.g. Engstrom, Persson, and Levander S, 1999), hopelessness and depression (e.g. Uncapher, Gallagher-Thompson, Osgood, and Bongar, 1998) or self-esteem (e.g. Overholser, Adams, Lehnert, and Brinkman, 1995). Other studies are concerned with interpersonal interactions (e.g. Zhang and Jin, 1998), psychoanalytic processes (e.g. Goldberg, 1999), life events (e.g. Cavanagh, Owens, and Johnstone, 1999), developmental factors (e.g. Van der Kolk, Perry, and Herman, 1991) etc. In the coherence model of suicide, the basic elements are the cognitive unit sets.

A cognitive unit set (CUS) is a collection of interrelated constructs, evaluations, beliefs, affects, or other cognitive units that focus on the same external or internal object or situational domain and can usefully be regarded as a whole. The meaning of a CUS depends on the associations between its component cognitive units and on the associations with other CUSs. A CUS has a positive pole or cognitive unit subset and a negative pole or cognitive unit subset. The direction of a CUS depends on the difference between the strength of its poles: the bigger the difference, the less ambiguous the direction. The cognitive units forming the positive pole and the cognitive units forming the negative pole negatively constrain one another. For example, in the case of ego CUS, cognitive units such as "I am a good person", "I am a successful person" etc. (positive pole) are negatively associated with cognitive units such as "I am a bad person", "I am a loser" etc. (negative pole). Two CUSs can be positively or negatively associated. If positively associated, the constraint between their positive poles and the constraint between their negative poles are positive, whereas the positive pole-negative pole constraints are negative. For instance, economic CUS is positively associated with ego CUS: positive cognitive units focused on the personal economic situation ("I’m richer than most people I know", "I have a decent financial situation" etc.) activate ego positive cognitive units and deactivate ego negative cognitive units. Economic negative cognitive units ("I’m poorer than most people I know", "My financial situation is very constraining" etc.) deactivate ego positive cognitive units and activate ego negative cognitive units. If two CUSs are negatively associated, the constraints between their positive poles and between their negative poles are negative, while the positive pole-negative pole constraints are positive.

The cognitive unit sets are represented by sets of units in a connectionist network. Two negatively connected nodes  - excitatory and inhibitory - represent a cognitive unit subset. A CUS is represented by two negatively connected cognitive unit subsets - one standing for the positive pole, the other for the negative pole. The excitatory and the inhibitory nodes of a subset are construed according to the rules specified by Shultz and Lepper (1996): inhibitory nodes have smaller maximum activation values (0.5) compared with excitatory nodes (1.0). However, in the coherence model of suicide two negatively connected nodes represents a subset of cognitions and not a particular cognition.

This representational schema tends to be more sensitive to the complexity and the ambiguity of human cognition. The fact that a person can hold simultaneously contradictory beliefs about the same object can be easily represented in the coherence model, by assigning similar initial activation values for the two poles of the CUS that focus on the respective object. Ambiguity plays an important role in suicide (Cosman, 1999; Firestone, 1997).

The CUSs are intra- and interrelated. Two CUSs can be positively or negatively connected (Figure 1a and 1b).
 
 

Figure 1 Positive (a) and negative (b) connections between two CUSs. Solid lines represent excitatory connections and dashed lines inhibitory connections. Big solid line rounded rectangles represent CUSs. Small solid line rounded rectangles represent positive poles; small dashed line rounded rectangles represent negative poles. The small circles stand for the excitatory (+) and the inhibitory (-) nodes of a pole.

If two CUSs are positively connected (Figure 1a), their positive poles are positively connected, their negative poles are also positively connected, while the positive poles-negative poles connections are negative. The situation is reversed for negatively connected CUSs (Figure 1b). A positive connection between two poles means that the links between their excitatory and between their inhibitory nodes have positive weights, while the links between their opposite nodes have negative weights. The same weights have reversed signs if the connection between two poles is negative.

As for the connections between the component cognitive units of a CUS, the excitatory and the inhibitory nodes of the two poles are negatively connected. The poles are also negatively connected. That is, negative connections link their excitatory and their inhibitory nodes, while the connections between the excitatory and the inhibitory nodes are positive.
 
 

The relevant CUSs and the relations between them

Most sociological studies of suicide explore the relations between suicide rates and various economic (e.g. income), professional (e.g. professional status), relational (e.g. divorce), cultural-religious (e.g. church membership), and personal factors (e.g. mental disorders). Common sense attributes suicide causes to similar categories. The coherence model of suicide assumes that these factors represent a natural categorization of the most important situational domains that affect suicide. The nine CUSs considered essential in the development of suicidal thoughts closely follow this categorization:
 

• Economic CUS
• Relational CUS
• Professional CUS
• Health CUS
• Ego CUS
• Suicidal thought CUS
• Life CUS
• Death CUS
• Suicide CUS

Factors like alcoholism, age, sex, religion etc., which are important in the classifications previously mentioned, are assumed to affect suicide indirectly, through the modification of the nine basic CUSs involved in the development of suicidal thoughts. For example, studying a general relation between religion and suicide rates can be useful in some studies. In the coherence model, the most plausible solution is to assume that there are several central attitudes that affect suicidal thoughts, in this case the way people think and feel about life (life CUS), death (death CUS), and suicide (suicide CUS), and that religion affects them and the relations between them. This focus on the manner individuals construe suicide related factors, although very rare, is not absent in the sociological literature (Stack, 2000; Douglas, 1967).

The nine CUSs constrain one another, with some constraints being more important (Figure 2). Consistent with the construction of suicide as escape from aversive self-awareness (Baumeister, 1990), the constraint between ego and suicidal thought CUSs has a central importance in the coherence model of suicide. It reflects the basic assumption that the most important source in the development of suicidal thoughts is a negative oriented ego CUS. The relation is statistical: the model does not exclude the possibility that, in some cases, positive ego cognitive units are associated with suicidal thoughts; it only considers that negative constraints prevail. This remark is relevant for the nature and the strength of the constraints between the other CUSs too.
 
 

Figure 2 The constraints between the nine basic CUSs. Bold lines represent the most important constraints. Solid lines represent positive constraints while dashed lines represent negative constraints. A complete representation of these sets and of the connections among them would result in a figure with 36 units and 630 connections.

The coherence model assumes that the way individuals think and feel about life, death, and suicide also affect suicidal thoughts. The connections between the suicidal thought CUS and the life, death, and suicide CUSs represent this assumption. Life CUS is negatively associated with suicidal thought CUS, while death and suicide CUSs are positively associated with suicidal thought CUS.

The relations of the ego CUS with the economic, relational, professional and health CUSs represent the assumption that the way individuals think and feel about themselves is strongly and positively related to the way they think and feel about their economic, relational, professional, and health situations.

The other connections reflect, for example, that a negative orientation of the death CUS will deactivate the positive pole of the suicide CUS and will activate its negative pole, a positive oriented health CUS will activate the positive pole of the life CUS and deactivate its negative pole, etc.
 
 

Process

The coherence model of suicide assumes that events activate the network. The activation spreads in parallel from the CUS that is initially activated by the event (it can be any CUS) to the other CUSs in the network and all the units are updated until the network settles down. For example, if a person reads a book where suicide is described as an honorable solution to various existential problems, the model supposes that this event activates suicide CUS (it does not explain how CUSs are activated by events). The activation spreads to the other CUSs according to the constraints depicted in Figure 2. The positive pole of suicide CUS activates the positive poles of death and suicidal thought CUSs and deactivates their negative poles, whereas the negative pole of suicide CUSs deactivates the positive poles of death and suicidal thought CUSs and activates their negative poles etc. There are 16 connections that work at the same time for two negatively or positively connected CUSs, plus the 6 connections between the component units of each CUS (Figure 1). The preexisting knowledge base, reflected by the nature and the strength of the connection weights and by the initial unit activation, constrains the spread of the activation. If the person reading the book has very strong and negatively oriented suicide and death CUSs, and very strong and positively oriented life, ego, economic etc. CUSs, the model predicts that is very unlikely this event will change the activation pattern. On the other hand, if the person has a negatively oriented health CUS, negatively oriented and strongly related economic, relational, professional, and ego CUS the result will depend on the strength and direction of suicide, suicidal thought, life, and death CUSs, and on the constraints between them.

The coherence optimization model deals with the automatic components of suicidal thoughts. Obviously, suicidal thoughts processes are not just automatic processes and suicide processes are not just suicidal thoughts processes (Retterstøl, 1993). However, suicidal thoughts have a central role in the suicidal process (Truant, O'Reilly and Donaldson, 1991). Individuals have a base of interrelated cognitive units that represent the starting point from which they integrate new pieces of information. This base being relatively stable (Mischel and Shoda, 1995), it is likely that at different moments the same individual will do the integration of new information in the same direction. To put it differently, if a person has suicidal thoughts today, it is likely that he or she will has the same thoughts tomorrow. Some persons will succeed to cope with their situation and avoid suicide, while other persons will not.
 

Social level

Individuals are not identical in respect to their CUS strength and direction and in respect to the strength and nature of the constraints between their CUSs, this cognitive variability being represented in the coherence model by different distributions of CUS activation and of connection weights. The model assumes that suicide rates depend on the CUS activation and connection weight distributions. It also assumes that these distributions are affected by different social factors (economic, cultural, religious etc.) and that the relation between these factors and CUS distributions is complex, the same social factor affecting many CUSs, and the same CUS being affected by many social factors. Suicidal rates are construed as summation, the simplest form of individual behavior aggregation (Boudon, 1997).
 

SIMULATIONS

The coherence model of suicide has been implemented in a computer program. The activation of excitatory units varies between 0 and 1, and the activation of inhibitory units varies between 0 and .5; negative connection weights vary between –1 and 0, and positive connection weights vary between 0 and 1. The program has a procedure that cycles the network until it settles down. Every cycle, units are randomly selected and updated. That is, the program first calculates the net input of the selected unit (equation 4), then applies the logistic function to the selected unit using as argument its net input (equation 3). The number of units selected per cycle is equal to the total number of units in the network. The procedure ends when units do no longer significantly change their activation from cycle to cycle. At this moment, the final activation of the suicidal thought CUS is recorded. This is the individual level.

The program also has a procedure that repeats the first one a large number of times (100,000). Every time the first procedure is called, initial unit activation and the connection weights are randomly attributed to the 36 units and 630 connections forming the network by a function that creates, for each unit and each connection, distributions with specific means. For instance, if the function construes for an excitatory node an activation distribution that has an average of .20, it randomly attributes to the respective unit 100,000 different values that range from 0 to 1, and average .20. If a CUS has a medium activation of its positive pole greater than the medium activation of its negative pole, not all the cases will have a positively directed CUS. Actually, in some of the 100,000 cases the activation of the negative pole will be greater than or equal to the activation of the positive pole. During the simulation, there are extremely small chances to record duplicate unit activation and connection weight patterns. This connectionist variability parallels people’s cognitive variability and offers a certain degree of psychological realism to the coherence model. This is the social level.
 

In order to get the final picture, after all 100,000 cases have been simulated, the program counts the number of times the suicidal thought CUS has had a positive final direction. Relying on personality (Mischel and Shoda, 1995; McCrae and Costa, 1995) and environment stability (Durkheim, 1897/1993), the model assumes that the final number gives an idea of the suicidal rates. Suicidal rates are supposed to be smaller because not all the people who have suicidal thoughts actually commit suicide: there are also suicidal attempts, microsuicidal behavior etc. The passage from intentions to behavioral performance requires behavioral scripts, procedural knowledge (Miller, Shoda and Hurley, 1996), and other cognitive units that are not implemented in the coherence model. However, the group of suicidal persons comes from the group of suicidal thought persons, and it seems reasonable to assume that variations in the number of suicidal thought persons are positively associated with variations in the number of suicidal acts.
 
 

Suicidal rates: constancy and variability

Durkheim started the construction of his theory from the fact that, usually, suicidal rates have similar values from year to year. According to the coherence model, suicidal rates are constant if the CUS distributions are stable. That is, if the way individuals construe their economic and relational situations, their ego etc., do not significantly change form year to year. The model assumes that these distributions are constant if the environment individuals live in and their personalities are stable.

In order to see how the connectionist program works with constant distributions, ten consecutive simulations have been performed using the distributions described in Table 1 and 2.
 
 
 

The suicidal thought, death, and suicide CUSs are negatively directed, while the ego and health CUSs are positively directed. The economic, relational, professional and life CUSs are also positively directed but no so strongly as ego and health CUSs. This activation pattern reflect the assumption that, in this virtual society, most people have negative attitudes toward death and suicide, positive attitudes toward their egos as well as toward their health, economic, relational, and professional situations, and do not have suicidal thoughts. There are strong relations between the ego and the economic, relational, professional, health, and suicidal thought CUSs, and between the suicidal thought, life, death, and suicide CUSs.

The results (Figure 3a) represent how many times the network has settled down with a positive direction of the suicidal thought CUS. Comparatively, in Figure 3b, is presented the situation of suicides in Denmark, using data offered by Durkheim (Durkheim, 1897/1993). The simulated results have small variations from year to year, even smaller than expected. This is happening because the simulations assumed an ideal situation: the CUS distributions are constant for a long period. It is more realistic to assume that this constancy is only relative and that actually, there are variations in these distributions from year to year and even during the same year (this can be a plausible explanation of the seasonal variances of suicides). However, even with perfectly constant distributions, it can be observed that there are fluctuations in the simulation graph that are attributable exclusively to hazard.
 

                                (a)                                                                              (b)

Figure 3 (a) Simulation data with constant unit activation and connection weight distributions and (b) suicides in Denmark, from 1841 to 1850 (data from Durkheim, 1897/1993). (a) is not a  simulation of (b), they are presented together only for illustrative purposes.

Suicide rates may undergo dramatic changes. For example, in the last 45 years suicide rates have increased by 60% worldwide (World Health Organization, 2000). The coherence model of suicide assumes that the modification of CUS activation and connection weight distributions affect suicidal rates. Some changes have been made in the distributions described in Table 1 and 2, in order to see how they will affect the simulation results. These modifications are reflected in Table 3 and 4. The first simulation has a less positively directed economic CUS (the positive pole is less activated and the negative pole is more activated), a trend that continued another three “years”. In the fifteenth consecutive simulation, the professional CUS medium direction also becomes less positive, while in the sixteenth simulation there are more negatively oriented suicidal thought, death, and suicide CUSs.
 
 

Finally, the connections between the ego and economic CUSs are also altered (Table 4). All other distributions are identical to those presented in Tables 1 and 2, for each simulation. The results (Figure 4) show that the modifications of cognitive unit set distributions also modify the total number of the suicidal thought CUSs with final positive activation.
 
 

Figure 4 simulation data: the modification of some CUS activation and connection weight distributions starting from the 11^th simulation, alter the number of times the suicidal thought CUS has a final positive direction.

The decrease of the medium positive direction of the economic CUS determines an increase in the number of suicidal thought CUSs with a final positive direction. The biggest number of suicidal thought CUSs with final positive direction is registered when the decrease of economic CUS medium positive direction is associated with the decrease of professional CUS medium positive direction. When the death and suicide CUSs become more negatively oriented the number of suicidal thought CUSs with a final positive direction decreases. This trend continues the next simulation, because the constraints between the economic and ego CUSs are altered.
 

Group differences

Researches on suicide rates show significant differences between groups of people. According to the coherence model, such differences are determined by significant group differences of CUS distributions. That is, some groups have different suicide rates compared with other groups because the people that form these groups construes differently the most important objects or situational domains affecting suicide.

There is a well-known difference between men and women suicide rates, male suicide rates, with extremely few exceptions, being bigger than female suicide rates (World Health Organization, 2000). In order to simulate such a result there must be made some psychological assumptions that will be reflected in the distributions of the CUS activation and of the connection weights used by the connectionist program.

First, in many cultures, the most important role for men is economic and professional success while for women the most important role is relational success (Stack, 2000). Consequently, men will have stronger constraints than women between the economic and ego CUSs, and between the professional and ego CUSs. Women will be less affected by a negative direction of the economic or professional CUSs, but also less protected by a positive direction of the same CUSs. The constraints between the relational and ego CUSs will be stronger for women and weaker for men. Compared with men, the female medium activation of the relational CUS is bigger, while the medium activation of economic and professional CUSs is smaller.

Stack’s review of sociological researches on suicide (Stack, 2000) mentions other important differences: women are more religious than men, have smaller alcohol abuse rates, and are also more open to professional help when facing psychological troubles etc. These results will be reflected in the coherence model by different activation means for life and suicide CUSs: women will have a more negative oriented suicide CUS and a more positive oriented life CUS.

Obviously, this pattern is not the only one possible. It would be also reasonable to expect differences between the death, health or ego CUSs of men and women. The different male/female suicidal rates ratios reported by various studies imply that cognitive variability may be even greater.

The distributions used in men-women simulation are illustrated in Tables 5 and 6.
 
 
 

Figure 5 shows that the simulation data indicate a constant difference between the numbers of times the two pattern of activation and connection distributions result in a final positive direction of the suicidal thought CUS, men having more suicidal thought CUSs with a final positive direction than women.

Figure 5 Simulation data: different distributions of CUSs and of the constraints between them for men and women result in a different number of suicidal thought CUSs with a final positive direction.

The coherence model of suicide approaches the differences between other social groups similarly: a) it assumes that differences in suicidal rates are determined by differences in the CUS distributions and in the distributions of the connections between them; b) it construes these distributions according to the actual state of knowledge about the relevant cognitive variables  of those groups.
 

DISCUSSION

The connectionist program that implements the coherence model of suicide successfully simulates suicidal rates constancy and variability, as well as suicidal rates group differences. These results, which capture important aspects of suicidal phenomenon, seem to confirm the idea that the same model can effectively approach both individual and social dimensions of suicide, without being constrained to reject or ignore one of them.

In order to explain suicidal rates, Durkheim used objective social forces that constrain individual behavior in the same way gravitation or other objective forces do. In his work, Durkheim mentioned a possible alternative to his own explanation, but rejected it. “Would it not be possible, he wrote, that the different incidents of private life, which are determinant causes of suicide should return regularly every year, in the same proportion? …Consequently, it would be natural to suppose that individuals, reaching a similar number of analogous situations would produce an equal number of voluntary deaths…Nevertheless, we know that although individual events precede suicides, they are not their real causes. There is no misery in life to determine someone to take his life if a supplemental predisposition of different nature would not exist. Therefore, the regularity with which favorable circumstances are repeated can not justify the regularity of suicide” (Durkheim, 1897/1993, p 248).

Although not specified by Durkheim, in his event regularity alternative hypothesis description there is an implicit assumption that humans have stable reactions to the events they experience. Similar reactions to similar events require a stable personality. In spite of this implicit assumption, Durkheim rather focuses on the variability of human reactions to experiences that seem to trigger suicide, and maintains that this variability demonstrate that individual factors cannot determine suicidal rates. According to him, the relative constancy of suicidal rates can be explained, “only by the permanent action of an impersonal cause situated above all particular causes”. Durkheim thought that collective currents or tendencies represent this cause.

The coherence model of suicide takes a different approach. It focuses on the information considered essential in the formation of suicidal thoughts, providing a schematic representation of this process. Individual factors, rejected by Durkheim because he assumed that they provide an illusory explanation, play an essential role here. The basic elements are the cognitive unit sets, which have a stable organization and are the object of a coherence optimization process. They are represented by sets of units in a constraint satisfaction network and are characterized by direction and strength. The psychological associations between them are represented by positive or negative connection weights. It is assumed that the events occurring in important life areas change the way individuals perceive their situation. If these changes have a negative direction (e.g. they become very dissatisfied about their relational situation) and individuals have a negative CUS pattern (e.g. a positive attitude toward suicide, negatively oriented ego CUS etc.) then they may develop suicidal thoughts. The model accepts that social factors explain suicidal rates, but this is a distal explanation and not a proximal one: social factors affect suicide rates because they affect the population’s distributions of the CUSs implied in the formation of suicidal thoughts. However, the model is concerned less with the relation between social factors and these distribution, assuming that the same factor may alter more CUSs and the same CUS may be altered by different social factors, and more with the way CUS distributions affect suicide rates. The constancy and the variability of the distributions of relevant cognitive variables implied in the development of suicidal thoughts explain the constancy and variability of suicidal rates. Group differences in the distributions of relevant cognitive variables explain group differences in suicidal rates.

Compared with other psychologically sensitive approaches of suicide, the coherence optimization approach offers a more precise specification of its basic units and process. In spite of ideas common to other approaches of suicide, the model has a clear and distinctive orientation, including suicide in the more general family of coherence optimization processes and offering a simple and natural explanation of both individual and social levels of suicide. The simulations performed by the connectionist program that implements the coherence model imply that this explanation may be plausible.
 
 

REFERENCES

1. Atkinson, J.M. (1978) Discovering Suicide. Studies in the Social Organization of Sudden Death, London: MacMillan.

2. Baumaister, R.F. (1990) “Suicide as Escape from Self”, Psychological Review, 97: 90-113.

3. Boudon, R. (1997) “Actiunea”, in R. Boudon (Ed) Tratat de Sociologie, Bucuresti: Humanitas.

4. Cavanagh, J. T. O., Owens, D. G. C. and Johnstone E. C. (1999) “Life events in suicide and undetermined death in south-east Scotland: a case-control study using the method of psychological autopsy”, in Social Psychiatry and Psychiatric Epidemiology, 12:645-650.

5. Cosman, D. (1999) Sinuciderea. Studiu in perspectiva biopsihosociala, Cluj-Napoca: Risoprint.

6. Douglas, J. D. (1967) The Social Meanings of Suicide, Princenton: Princenton Universitary Press.

7. Dumitrescu, D. and Costin, H. (1995) Retelele neuronale. Teorie si aplicatii, Bucuresti: Teora.

8. Durkheim, E. (1993) Despre sinucidere, Iasi: Institutul European.

9. Engstrom G, Persson B and Levander S (1999) “Temperament traits in male suicide attempters and violent offenders”, in European Psychiatry 14:278-83.

10. Firestone, W.R. (1997) Suicide and The Inner Voice, Thoushand Oaks: Sage Publication.

11. C Goldberg (1999) “A psychoanalysis of love: the patient with deadly longing”, in American Journal of Psychotherapy, 53:437-51.

12. Henri, A.F and Short, J.F. (1954) Suicide and Homicide, New York: The Free Press.

13. Holyoak, K. J. and Thagard, P. (1997) “The Analogical Mind”, American Psychologist. 52: 35-44.

14. Kunda, Z. and Thagard, P. (1996) “Forming Impressions from Stereotypes, Traits and Behaviors: A Parallel-Constraint-Satisfaction Theory”, Psychological Review: 103:284-308.

15. Mc Crae, R.R. and Costa, P.T. (1995) “Traits Explanation in Personality Psychology”, European Journal of Personality, 9:231-252.

16. Miller, S.M., Shoda, Y., Hurley, K. (1996) “Applying Cognitive-Social Theory to Health-Protective Behavior: Breast Self-Examination in Cancer Screening”, Psychological Bulletin, 119: 70-93.

17. Mischel, W. and Shoda, Y. (1995) “A Cognitive-Affective System Theory of Personality: Reconceptualizing Situations, Dispozitions, Dynamics and Invariance in Personality”, Psychological Review, 102: 246-268.

18. Overholser JC, Adams DM, Lehnert KL and Brinkman DC (1995) “Self-esteem deficits and suicidal tendencies among adolescents”, in Journal of the American Academy of Child&Adolescent Psychiatry 34:919-28.

19. Retterstøl, N. (1993) Suicide. A European Perspective, New York: Cambridge University Press.

20. Shultz, T.R. and Lepper, M.R. (1996) “Cognitive Dissonance Reduction as Constraint Satisfaction”, Psychological Review, 103: 219-240.

21. Stack, S. (2000) “Suicide: a 15-year review of the sociological literature. Part I: cultural and economic factors”, Suicide and Life Threatening Behavior, 30:145-62.

22. Thagard, P., and Kunda, Z. (1998), “Making sense of people: Coherence mechanisms”, in S. J. Read & L. C. Miller (Eds.), Connectionist models of social reasoning and social behavior, Hillsdale, NJ: Erlbaum.

23. Thagard, P. and Verbeurgt, K. (1998) “Coherence as constraint satisfaction”, Cognitive Science, 22:1-24.

24. Thagard, P., and Millgram, E. (1995) “Inference to the best plan: A coherence theory of decision” in D. Leake and A. Ram (Eds.), Goal-directed learning. Cambridge, MA: MIT Press.

25. Toderean, G., Costeiu, M. and Giurgiu, M. (1995) Retele neuronale artificiale, Cluj-Napoca: Editura Albastra.

26. Truant G. S., O’Reilly R. and Donaldson L. (1991), “How psychiatrists weigh risk factors when assessing suicide risk”, Suicide and Life Threatening Behavior, 21:106-114.

27. Uncapher H, Gallagher-Thompson D, Osgood NJ and Bongar B (1998) “Hopelessness and suicidal ideation in older adults”, in Gerontologist 38:62-70.

28. Van der Kolk, B.A., Perry, J.C. and Herman, J.L. (1991) “Childhood Origins of Self Destructive Behavior”, American Journal of Psychiatry, 148:1665-1671.

29. World Health Organization (2000), Suicide Prevention Program, [On-line], Available: http://www.who.int/mental_health/Suicide/index.html