1.1. Explanation

The clinical literature on human filicide lacks a comprehensive explanation. Prevalent professional accounts of filicide emphasize socioeconomic stressors such as poverty and unemployment (e.g., Belsky, 1993, Finkelhor, 1997, Goetting, 1988), mental illness (Bourget & Gagné, 2002, Goetting, 1988, Marleau et al., 1999, Resnick, 1969, Stroud, 1996, Stroud & Pritchard, 2001, Wilczynski, 1997, Xie & Yamagami, 1995), and marital disharmony (Adinkrah, 2003, Stanton & Simpson, 2002, Somander & Rammer, 1991). In a study of laypeople's explanations of child maltreatment, the prevalent opinions about etiology matched those advanced by experts in the field: poverty and family instability, substance abuse, stressors, moral ignorance, and individual pathology (Korbin, Coulton, Lindstrom-Ufuti, & Spilsbury, 2000). These explanations are insufficiently constrained; how poverty or family breakup induces some parents, but not most, to kill their children is unspecified, for example.

Those unfamiliar with the extensive literature on parent–offspring conflicts across species might wonder, “How could even a slight tendency to kill offspring ever be the result of Darwinian selection?” In mammalian species especially, high levels of parental solicitude represent an obvious and important aspect of reproductive fitness. Within that general tendency, however, a parent's reproductive interests are not isomorphic with those of each offspring. Extensive work (Hrdy, 1979, Hrdy, 1999, Scrimshaw, 1984) has identified five possible bases (the first four of which represent aspects of adaptive function) for offspring killing by parents: (a) sexual selection—reproductive opportunity produced by killing the offspring of another (e.g., killing a prior male's offspring, thereby hastening estrus); (b) parental manipulation—killing that directly improves the parents' reproductive fitness (e.g., killing newborns that would draw resources away from older or future offspring); (c) direct resource competition; (d) resource exploitation (e.g., cannibalism); and (e) pathology. Lay interpretation of human filicide generally implies only the last nonadaptive basis (Hrdy, 1999), but natural selection points to the first two of these as relevant to all filicides, including that by humans.

Among humans, data showing that filicide is associated with circumstances related to net ancestral reproductive fitness would argue for interpretations reflecting manipulation or sexual selection. These accounts depend on selectionist concepts of inclusive reproductive fitness and asymmetric parental investment. Selectionist accounts hypothesize that male reproductive fitness has been associated with proprietary behavior towards their mates. That is, some degree of coercion by men towards women and some degree of suspicion regarding their offspring is to be expected on the grounds that ancestral males who engaged in some form of coercion, threat, and even force to discourage their mates from sexual behavior with other men would be less likely to raise offspring who were not their own and, thus, were more reproductively successful. Threats that the perpetrator was demonstrably willing and able to carry out (in contrast to deliberate bluffing) could be expected to be the most potent of all (Daly & Wilson, 1988a, Daly & Wilson, 1994, Wilson et al., 1995). The result is that human male psychology has an evolutionarily based tendency to regard the sexual alienation of a spouse as a catastrophic loss that is to be resisted via aggressive high-stakes tactics that usually succeed but occasionally backfire, resulting in loss of reproductive fitness. As such, an apparently maladaptive behavior could “represent the tail of some motivational distribution” (Daly & Wilson, 1998a, p. 443). Media reports tell us that a father can be so irrationally distraught or angered by his wife's leaving him for another man that he kills her and her children. The selectionist approach both affords a source for such an extreme emotional reaction and explains why such lethal behavior is so uncommon in response to even the most extreme of life's other tribulations, or among mothers under any circumstances.

Similarly, filicide by stepparents is unlikely to carry direct adaptive advantages (Forbes, 2005), but the clearest example of parents' reproductive interests not coinciding with the child's occurs in stepparenting relationships. Many stepparent relationships are loving; indeed, on selectionist grounds, prospective partners can be expected to exhibit some stepparental solicitude as a hard-to-fake aspect of mating effort. Nevertheless, selectionist work on infanticide (e.g., Hrdy, 1999, McCleary & Chew, 2002, Turke, 1996) demonstrates that, in a world of limited resources, stepparental investment entails greater reproductive cost (and smaller reproductive benefit) than genetic–parental solicitude. Despite its widespread representation in folklore, the association between maltreatment of children and stepparenthood has only rather recently been empirically examined (Wilson, Daly, & Weghorst, 1980). Research motivated by selectionist thinking found that stepparents, compared with genetic parents, represent a much greater risk of violence and death to children, independent of such other risk factors as parental age and poverty (Daly & Wilson, 1998a). The greater tendency of stepfathers to beat children to death in a rage (Daly & Wilson, 1988a, Daly & Wilson, 1994, Weekes-Shackelford & Shackelford, 2004) suggests a failure of solicitude (as distinct from a direct adaptive advantage to homicidal behavior) compared to comparatively painless planned filicides by gunshot or asphyxiation by genetic parents (Weekes-Shackelford & Shackelford, 2004).

There has been very little research on stepmaternal filicide. Because the risks to reproductive success associated with indiscriminant stepparental solicitude are greater for women (who experience much less variability in reproductive success) than for men, women would be expected to exhibit even greater differences in levels of parental solicitude as a function of their status as either genetic mothers or stepmothers. Furthermore, human fatherhood entails more parental uncertainly than motherhood (due to extrapair mating, concealed ovulation, and internal gestation) such that parental solicitude would be expected to be more similar among fathers and stepfathers than between mothers and stepmothers.

Perhaps the most challenging case for laypeople to comprehend—the mother who kills her own child—is most likely to be explained from a selectionist account as having had direct adaptive benefits. Mothers are obliged to invest much in each offspring, through gestation and lactation, and the consequent sacrifice of opportunities to produce other offspring. Thus, they stand to lose much by investing in offspring who would not thrive as well as might later offspring. Occasions on which the mother of an infant elects to let or to cause the child to die are expected to be most common when the child is an infant (relatively few resources have already been expended), the mother is young (she has more prospects for later successful reproduction), her resources are scarce, paternal support is undependable for any reason, or the infant has obvious health problems. These would all be circumstances in which women's ancestral reproductive fitness might have been enhanced by parental manipulation—killing a child in order to devote more resources to older offspring or to delay reproduction until more favorable circumstances are likely to prevail. Thus, this selectionist account can explain the well-established finding that maternal filicides exhibit a greater inverse relationship with age than child homicides perpetrated by anyone else. The killing of healthy children by genetic mothers after their first year of life is sufficiently rare that one could hypothesize that it is primarily a true pathology or disorder (Daly & Wilson, 1988a). Following Wakefield (1992), we define a “disorder” as a harmful dysfunction (i.e., the failure of a mechanism to perform the function for which it was designed by natural selection). Compared to all other perpetrators, genetic mothers who kill noninfants are more likely to be diagnosed with serious mental disorders and to be excused on account of insanity.

The explanatory utility of a selectionist account lies on its ability to make sense of the overall pattern of child homicide by humans, including comparisons between male and female perpetrators, between genetic and stepparental perpetrators, and even between maternal perpetrators of infants versus maternal perpetrators of older children (Janson & van Schaik, 2000). It does not simply account for Cinderella's treatment at the hands of her stepmother (Daly & Wilson, 1998b), but uniquely predicts how she would be treated differently by her mother, father, or stepfather, or indeed by her stepmother had there been no stepsisters. The issue addressed by the present research is the accommodation of proximal (and often apparently pathological) factors related to individual findings pertaining to filicide by more distal selectionist hypotheses.

1.2. The present study

There is limited research on children killed by stepmothers, and little of the research on human filicide has systematically compared all four categories of perpetrators (genetic mothers, genetic fathers, stepmothers, and stepfathers) and cases in which children were killed by nonfamilial adults. As well, few previous studies have examined the circumstances of the offenses in detail, in part because research has relied on general registers containing limited data on the offense. The present study employed detailed materials gathered by police investigators in the form of a standardized investigative database, permitting us to compare the manner of death of children killed by genetic parents and stepparents with that by nonfamilial adults, and to examine three of the hypothetical bases for filicide suggested by selectionist accounts. Specifically, we hypothesized that pathological factors would most characterize filicide by genetic mothers of older victims because those offenses are least likely to serve the perpetrator's reproductive fitness. Thus, consistent with prior research, we expected that mental illness and legal findings of insanity would be most common among genetic mothers killing older children.

We hypothesized that parental manipulation would, however, characterize many other filicides by genetic mothers, especially young mothers of infant victims, or unhealthy older victims. Thus, and consistent with prior research, we expected that sadness or despair, rather than hostility, would be most common among genetic mothers killing infants and that the manner of causing death would be comparatively painless rather than implying anger or rage. As well, some mental illnesses (especially depression) were expected to be associated with the lack of personal and material resources. We also hypothesized that filicide by fathers, especially genetic fathers, would be characterized by coercive tactics aimed at controlling the reproductive behavior of mates. We regarded this form of filicide as a behavioral “tail end” of parental manipulation. Thus, and consistent with previous research, prior marital discord and conflict, and uxoricide, familicide, and suicide would be most common among fathers, and especially among genetic fathers.

Among stepparents, and especially stepmothers, we hypothesized that evidence of weaker parental solicitude and resource competition would be most common. Thus, and consistent with prior research, stepparents would represent a greater risk of filicide than would genetic parents. In addition, denial of resources (i.e., neglect) and reduced parental solicitude in the form of prior abuse and overt hostility, anger, rage, and beating to death would be most common among filicides by stepparents. In particular, we hypothesized that the reproductive costs of stepparental solitude are greater for women than for men; thus (novel to this study), previous neglect and abuse, anger/rage, and beating to death would be most common among filicides by stepmothers. We further hypothesized that the effect of direct resource competition would be heightened by the presence of offspring genetically related to the perpetrator; thus (novel to this study), genetic offspring would be associated with worse previous abuse and neglect by stepparents.

2.1. Subjects and procedure

Cases were drawn from the Violent Crime Linkage Analysis System (ViCLAS), a national police database of serious crimes designed to assist police services across Canada to identify similar cases being investigated across jurisdictions (Collins, Johnson, Choy, Davidson, & MacKay, 1998). The system began its operation in 1996 as an optional police reporting system and became mandatory in February 1997 for all qualifying offenses, including murder and attempted murder. Some cases that occurred prior to 1996 were subsequently entered into ViCLAS, and no cases have been purged. There is an associated audit process to ensure that all police services comply with the submission of qualifying cases. The database is maintained by the Royal Canadian Mounted Police, a national police service with recording centers across the country, including the one used for the present study in the Behavioral Sciences Section at the Ontario Provincial Police headquarters. Investigating officers are trained to complete ViCLAS reports consisting of over 150 items. Investigators are trained to be complete and comprehensive regarding offense details, and to attend carefully to the temporal sequence of events in a narrative summary. Investigators submit ViCLAS reports during the course of investigations and are not restricted to cases resulting in criminal charges. It was not possible for us to evaluate the reliability of the data in ViCLAS reports, but we did conduct an evaluation of the interrater reliability of coding in the present research as described below.

The details reported about each case reflect the purpose of the system as an aid to crime solving. Nevertheless, many were also relevant to our hypotheses, including detailed descriptions of the victims' age, sex, educational status, premortem health status, and living situation. Investigators reported extensively on perpetrator personal factors and history including physical description, occupation, living situation, marital status, lifestyle, psychological and criminal history, information, and specific offense and post offense behavior. They also recorded details of the fatal offense: location, perpetrator–victim contact, sexual and nonsexual violence, type and method of injury, weapons employed, and the discovery of the body. From this database, we extracted all Canadian cases involving the homicide of a person under the age of 12 years where the perpetrator was believed to be known by police, even if not charged or convicted. This yielded 385 fatal cases prior to 2003.

A limitation inherent in this research is that the case must have been identified as a suspected homicide for it to be recorded. Such ascertainment is a limitation of almost all research studies on forensic samples, but in the present study, ascertainment might have been confounded with perpetrator category. For example, parents who caused the deaths of very young children might not be identified if they disguised the offense as death by accident or by natural causes (e.g., sudden infant death syndrome), which is not required to be recorded in ViCLAS. On the other hand, it might have been more difficult for killers of older children, who we hypothesize to have disproportionately been fathers, stepparents, and extrafamilial perpetrators, to have disguised their crimes. Although not a complete solution, we note that most Canadian jurisdictions apply additional levels of scrutiny to cases of child death, including those by apparent accident or natural causes. For example, in Ontario (Canada's most populous province), a Pediatric Death Review Committee, which reviews all cases of child death that were medically complex, was created in 1991 as part of the Office of the Chief Coroner. In 1997, it was decided that all child deaths (at any age) involving an open child protective service file also be reviewed. Furthermore, cases where the family or caregivers express any concern are also reviewed. In 1997, a new subcommittee was created and, since then, all child deaths under the age of 2 years have been also reviewed by an independent Deaths Under Two Committee, which has the ability to refer cases to the Pediatric Death Review Committee. The purpose of all these reviews is to independently scrutinize every death under the age of 2 years (no matter what the apparent cause) and any other complex, suspicious, or incompletely explained death of a child in order to ensure a complete police investigation and to order a coroner's inquest, when required. In addition, to help examine the potential scope of any ascertainment bias in the present study, we coded whether the identified perpetrator made any efforts to disguise the offense. We return to this potential limitation in the Results and Discussion section.

For the present study, research assistants, who were blinded to the study hypotheses but had several years' experience in coding similar police data, translated details directly from the database and coded several variables from narrative summaries. A separate record was derived for each child fatality whether or not more than one child in a family died on a single occasion. A few (n=7) cases in which the perpetrator was a nonparental family member were dropped, leaving a total of 378 cases that met the inclusion criteria. In 18 cases in which a genetic parental perpetrator had an accomplice, seven were nonparents or unknown, eight were the other genetic parent (these were assigned to the genetic mother category), and three were the stepmother and genetic father (assigned to the stepmother category); subsidiary analyses indicated that assigning these to the respective paternal categories resulted in the same results reported. There were no cases in which the victim was recorded as having been adopted as an infant (i.e., under the age of 1 year); so, if any such adoptions occurred, they would have been assigned to a genetic parent category. One older child was recorded as having been killed by an adoptive father and was assigned to the “stepfather” category. Summary data for the qualifying cases are shown in Table 1, Table 2.

Table 1. Case characteristics: means or dichotomous percentages ±95% confidence limits

	Variable	Genetic relationship		Nongenetic relationship
		Mother (n=111)	Father (n=86)	Mother (n=16)	Father (n=62)	Nonfamily (n=103)
	1. Victim's sex (% male)	51±9	57±10	50±25	56±12	41±9
	2. Perpetrator's age in years	27±2	34±2	27±4	27±2	26±2
	3. Victim's age in months	26±7	49±10	52±19	33±8	72±10
	4. Victim lived with single parent (%)	22±7	33±10	0	12±8	9±5
	5. Victim lived with step or half sibling (%)	1±2	1±2	38±24	16±9	13±5
	6. Homicide had a sexual element (%)	0	0	0	19±10	50±10
	7. Perpetrator attempted to deny offense (%)	81±8	78±9	81±19	76±11	66±9
	8. Perpetrator “discovered” the body (%)	13±6	14±7	25±21	13±8	5±4
	9. Perpetrator committed suicide (%)	11±6	40±10	6±12	6±6	9±6
	10. Perpetrator reported to be suicidal (%)	37±9	48±10	13±16	15±9	12±6
	11. Perpetrator's criminal history score	0.89±0.46	3.7±1.8	2.3±3.6	13.4±7.1	13.3±5.3
	12. Perpetrator's sentence (years; life=50 years)	10.4±4.3	10.1±3.5	15.4±11.3	14.8±3.4	18.4±2.6
	13. Perpetrator acquitted due to insanity (%)	26±8	4±4	9±14	0	9±6

As explained in the text, any mean or percentage lying beyond the confidence limits of another is a statistically significant difference.

Table 2. Comparisons for variables involving familial perpetrator–victim relationships: means or dichotomous percentages ±95% confidence limits

	Variable	Genetic relationship		Nongenetic relationship
		Mother	Father	Mother	Father
	1. Beating death score	0.61±0.21	0.59±0.17	1.6±1.0	1.3±.30
	2. Perpetrator used weapon or instrument (%)	78±13	69±16	50±28	39±16
	3. Perpetrator's anger/rage score	0.25±0.12	0.58±0.17	0.63±0.35	0.85±0.15
	4. Victim's abuse/neglect score	0.15±0.13	0.03±0.04	1.0±0.9	0.16±0.12
	5. Perpetrator previously injured victim (%)	13±6	10±6	38±19	18±9
	6. Severity of perpetrator's prior injury to victim	0.54±0.28	0.36±0.25	1.5±1.1	0.65±0.39
	7. Marital conflict/problems score	0.18±0.11	0.73±0.20	0.31±0.32	0.21±0.12
	8. “Shaken baby syndrome” (%)	8±5	19±7	13±18	26±7
	9. Perpetrator's serious mental problems score	0.97±0.20	0.43±0.16	0.18±0.30	0.27±0.15
	10. Victim was subject of official concern (%)	6±5	2±4	13±18	5±7
	11. Perpetrator's substance abuse score	0.27±0.14	0.41±0.16	0.44±0.68	1.1±0.4
	12. Perpetrator intoxicated at the time of offense (%)	2±2	8±4	13±18	15±8

As explained in the text, any mean or percentage lying beyond the confidence limits of another is a statistically significant difference.

Some of the variables shown in the tables require elaboration. An offense was rated as having a sexual element (Table 1, Row 6) when the perpetrator sexually assaulted the victim or made some attempt to do so. Perpetrators who “discovered” the body attempted to avoid apprehension by pretending to have found an already dead victim (Table 1, Row 8). A perpetrator's criminal history for violent and nonviolent offenses (Table 1, Row 11) was rated using an expanded version of a well-validated scale capturing the frequency and severity of criminal conduct (the Cormier–Lang scale; Quinsey, Harris, Rice, & Cormier, 2006). Beating death score (Table 2, Row 1) was the number of distinct ways reflecting the sole use of the perpetrator's body to cause death (punching, kicking, stomping, manual strangulation, etc). Weapon or instrument (Table 2, Row 2) referred to the use of any implement(s) or agent(s) to cause death (stabbing, shooting, burning, ligature strangulation, drowning, poison, etc.). Anger/rage score (Table 2, Row 3) was the number of distinct terms reflecting a perpetrator's anger (revenge, anger, and rage) that was noted by investigators as a motive. Abuse/neglect score (Table 2, Row 4) reflected the number of distinct terms (e.g., malnutrition, lack of appropriate care, and failure to provide adequate shelter or clothing) noted by police about fatal offenses or previous incidents, plus injury due to previous physical abuse. The severity of prior injuries (Table 2, Row 6) was rated on a 5-point scale ranging from 0=none to 6=life-threatening injuries or those resulting in disability. Marital conflict/problems score (Table 2, Row 7) was the number of distinct terms (e.g., sexual jealousy, marital separation, perpetrator or spouse with extramarital sexual relationship, child custody or access dispute, etc.) noted by police investigators as a relevant circumstance or motive. Serious mental problems score (Table 2, Row 9) was the number of distinct terms (e.g., depression, psychosis, delusion, hallucination, and extreme stress) used to describe the perpetrator's condition. The victim was rated as having been the subject of prior official concern (Table 2, Row 10) if the police investigators noted any report to formal child protective service agencies by anyone. Substance abuse score (Table 2, Row 11) was the number of distinct terms (e.g., alcoholic, drug addiction, intoxication, drug dealer, solvent abuse, etc.) police investigators used to characterize the lifestyle of the perpetrator.

The reliability of coding was assessed by having two researchers independently code 20 randomly selected cases on all variables. Items whose interrater reliability correlation coefficients did not exceed .80 were dropped, except for the dichotomous items “death by beating,” “death by suffocation,” and “death by gunshot,” each of which yielded κ=0.643 and was combined into composite variables, as described in the previous paragraph.

2.2. Parent-related denominators

We were able to find few Canadian data about children living with genetic (or adoptive) parents versus children living with stepparents (North American census agencies do not ask about this). According to one source, in 1995, 9% of Canadian children under 12 years who were living with a parent lived in a stepfamily (http://www.divorcemag.com/statistics/statsCAN.shtm, retrieved on January 10, 2006), and 16% lived with a single parent. Slep and O'Leary (2005) reported that, among 453 representatively sampled couples (95% married) with 3- to 7-year-old children, 99% of the women were the genetic mother and 95% of the men were the genetic father. Assuming that all children living in a family lived with at least one genetic parent (i.e., ignoring the small proportion living with adoptive parents or with a stepparent only), Canadian children were at least 10 times as likely to live with genetic parents as with stepparents. Other sources indicated that living with a stepfather is at least 10 times as common as living with a stepmother (Weekes-Shackelford & Shackelford, 2004). US data indicate that at least 2% of children are adopted (www.adoptioninstitute.org/FactOverview.html, retrieved on February 8, 2006; http://www.geocities.com/capitolhill/9606/statistic.html, retrieved on February 8, 2006; Stolley, 1993) and that over a third of adoptions are by stepparents. Thus, we expect two or three cases in the present sample in which a child was erroneously coded as a genetic offspring. However, this would have had the effect of working against the present study hypothesis.

4.1. Prevention and future research

An intriguing result pertained to high levels of criminal history among extrafamilial and stepfather perpetrators. The age of steppaternal perpetrators' victims was also unexpectedly young; in this group, there was an inverse correlation between the victim's age and the perpetrator's criminal history score (r=−.13, ns). We speculate that some stepfathers were psychopaths (i.e., a condition marked by extensive criminality; Harris et al., 2001, Lalumière et al., 2005, Lalumière et al., 2001) who “joined” a family to secure victims or whose filicides exhibited remnants of sexual selection (enhancing reproductive capacity by killing the offspring of prior males) as an aspect of general mating effort. Elsewhere, we have provided evidence suggesting that psychopathic men do not suffer from a true disorder but, rather, that they have been designed by natural selection to pursue a life history strategy different from that of other men—one based on high mating effort and sexual coercion, cheating, impulsivity, and callousness. Adding to this characterization, we predict that psychopaths would exhibit especially low parental solicitude and that men who kill stepchildren are especially likely to be psychopaths.

In this context, we wondered whether the high-stakes life strategy associated with psychopathy was associated with the high-stakes tactics characteristic of familicide. Comparing fathers who killed children only with those who also committed uxoricide led to a surprise—the familicide perpetrators had less serious criminal and antisocial histories [mean=2.11 (S.D.=4.95) vs. 12.64 (S.D.=27.77), t(190)=1.66, p<.10]. Criminal history is not an ideal measure of psychopathy, and future research could specifically assess the construct. Nevertheless, it appeared that the callous and impulsive lifestyle characteristic of psychopathy does not include the extreme emotional reactivity associated with familicide.

Previous research suggests that men who commit uxoricide yield high scores on actuarial assessments for the general risk of domestic violence (Hilton & Harris, 2005, Hilton et al., 2004), and that many paternal filicides are occasioned by domestic conflict and marital discord. However, because of their general lack of criminal history, the present results suggested that genetic fathers who also kill their children might not exhibit high scores on such standardized assessments for domestic violence. Future research might profitably examine the perpetrators of familicide in more detail with a view to enhancing the validity of available risk assessments, especially with potential predictors informed by selectionist hypotheses. Moreover, another fruitful avenue for subsequent research could be the application of the existing methods of psychological test construction to the development of formal risk assessments for use by professionals in cases of child abuse and neglect, especially research in which variables informed by selectionist thinking are entertained as potential empirical predictors of recidivism. Present and previous findings indicate that the youngest children are most at risk, especially from genetic mothers, and that such filicides are often occasioned by lack of material and personal resources, attempts to hide the birth, and mental illness, especially as children get older. These results point to the importance of prenatal and postnatal care for expectant and new mothers, and mental health care for all.

4.2. Final comment on explanation

The present findings were remarkably consistent with the prior research and selectionist analysis provided by Daly and Wilson (1984, 1988b) two decades earlier. Regarding the five possible bases for child killing, which were discussed in the Introduction, the present findings are most consistent with human filicide as reflecting various forms of parental manipulation (parental actions designed to seize control of reproduction by affecting resource allocation among offspring or by affecting the reproductive behavior of mates). Rarer cases of filicide by genetic mothers of older children do imply a role for pathology. We also note that some aspects of filicide by stepfathers resemble sexually selected infanticide by nonhuman males (Hrdy, 1999). More generally, we note that much apparently aberrant human conduct can be usefully understood as aspects of intrasex (e.g., sexually selected filicide) or intersex (e.g., parental manipulation) reproductive competition (Lalumière et al., 2005). A research challenge is empirically disentangling these two fundamental sources of explanatory power.