Childhood Unintentional Injuries: Factors Predicting Injury
Risk Among Preschoolers
Janet Abboud Dal Santo, DRPH, Robert M. Goodman, PHD, Deborah Glik, SCD,
and Kirby Jackson, BA
University of North Carolina at Chapel Hill (Dal Santo), University of Pittsburgh (Goodman),
University of California at Los Angeles (Glik), University of South Carolina at Columbia (Jackson)
Objective To examine the relationships between maternal perceptions of risk, stress, social
support, safety-proofing behaviors, supervision practices and unintentional injuries to children
under 5 years old. Methods Household interviews were conducted with 159 mothers who
had a preschool-age child. The secondary data were part of a population-based study that
collected self-report data and home observational data. Diaries were used for collecting
prospective injury data. Results White children whose mothers were unemployed and
whose homes needed repair were reported to be at higher injury risk than other children.
Predicting a higher injury risk were children’s behavioral characteristics as well as their being
older than 2.5 years. Maternal social support, stress, and coping variables were not related to
injury risk. Maternal perceptions of risk variables interacted with maternal safety behavior
variables when predicting injury risk. Conclusions Childhood injuries are predicted by
a set of interrelated sociodemographic, cognitive, behavioral, and child-related factors.
Key words unintentional injury; preschool children; prevention; psychosocial variables.
Unintentional injuries have replaced infectious diseases
as the most serious public health problem of children in
the industrialized world today. In the United States more
children between the ages of 1 and 4 die annually from
unintentional injuries than from all childhood diseases
combined (Centers for Disease Control and Prevention,
2001; Rodriguez & Brown, 1990). For every childhood
death caused by injury, there are approximately 18
hospitalizations, 233 emergency department visits, many
more visits to medical facilities, and a much larger
number of home-treated injuries (Grossman, 2000). The
impact of childhood injury is immense in terms of the
direct financial cost and the enormous emotional toll of
death and disability. For children aged 1 to 5, deaths
from unintentional injuries accounted for over one-third
of the 5,251 total deaths of children in this age group in
1998. Motor-vehicle crashes, drowning, fires, and burns
were the leading causes of death of preschool children
(National Safety Council, 2001). Injuries from falls and
poisonings are not major causes of death in this age
group, but they are quite prevalent and cause many
nonfatal and disabling injuries for young children
(Centers for Disease Control and Prevention, 2001).
Most unintentional injuries to children under 5 years of
age occur in and around the home (Reading, Langford,
Haynes, & Lovett, 1999).
Although the epidemiological literature has identified behaviors of both children and their parents as risk
factors for childhood injury, there is a dearth of research
on behavioral components of injury risk (Grossman &
Rivara, 1992; Peterson & Stern, 1997). Few studies
address associations between parental perceptions of
injury risk, parental safety behaviors, and injury risk to
children. Matheny (1986, 1987) studied patterns of
injury among 1 to 3-year-old children. Children whose
mothers were more stressed and less educated, and
Correspondence concerning this article should be addressed to Janet Abboud Dal Santo, DrPH, Injury Prevention
Research Center, University of North Carolina at Chapel Hill, CB #7505, Chapel Hill, North Carolina, 27599-7505;
e-mail: [email protected].
Journal of Pediatric Psychology 29(4) pp. 273–283, 2004
DOI: 10.1093/jpepsy/jsh029
Journal of Pediatric Psychology vol. 29 no. 4 Q Society of Pediatric Psychology 2004; all rights reserved.
274
Dal Santo, Goodman, Glik, and Jackson
whose homes were noisier and had more safety hazards,
were perceived by their mothers to be hard to manage
and aggressive, thus suffered higher rates of injury that
required medical attention. Significant associations were
reported between perceptions of risk and safety behaviors (Glik, Kronenfeld, & Jackson, 1993; Greaves, Glik,
Kronenfeld, & Jackson, 1994). Garling and Garling
(1993) investigated whether mothers believed that
supervision reduced the risk of unintentional injury to
children 1 to 3 years of age. Supervision reduced
perceived risk more among mothers of younger children
and in rooms perceived to be more dangerous. Peterson’s
research focused on injury prevention at the family level
and stressed the important role of adult supervision of
young children in risky situations (Peterson & Stern,
1997). While previous studies have examined parental
behaviors as outcome measures (Glik et al., 1993;
Greaves et al., 1994) this study extends previous work
by examining associations between maternal social,
cognitive, child-related, and behavioral variables and
child-injury outcomes.
The current study focuses on near miss and minor
childhood injuries rather than on severe or fatal injuries.
Injury events were studied prospectively in families as
they occurred, rather than retrospectively; thus, many of
the injuries were not severe. A unique feature of this
study was the use of diaries among mothers for collecting
prospective injury data. This method avoids the pitfalls
of recall bias associated with the use of a retrospective
research design (Harel et al. 1994; Verbrugge, 1980).
Having parents use diary records to report their
preschool children’s near miss and minor injuries has
been an accurate and reliable method of data collection
in a number of studies (e.g., Fonseca, Victora, Halpern,
Lima, & Barros, 2002; Marsh & Kendrick, 1999).
Concepts from two well-known psychosocial models were embedded in risk perception constructs that
were used to predict behavioral and injury outcomes.
The two dimensions of susceptibility and severity of
threat from the Health Belief Model (Janz & Becker,
1984) were used to construct the scale of Perceptions of
Risk of Injury (Glik, Greaves, Kronenfeld, & Jackson,
1992; Glik, Kronenfeld, & Jackson, 1991; Glik et al.,
1993). Also, the dimensions of likelihood and seriousness from Behavioral Decision Theory were used to
construct scales of Perceptions of Risk of Hazard (Slovic,
Fischhoff, & Lichtenstein, 1978, 1982). Concepts from
these models as well as findings from previous research
informed selection of exogenous and endogenous
variables in the predictive model that were deemed
relevant for the prevention of childhood unintentional
injuries (Kronenfeld & Glik, 1990). Endogenous determinants include cognitive variables such as perceived
risks of hazards and injuries, perceived child behavioral
characteristics, parental safety behaviors, and injury
history. Exogenous determinants in the model comprise
sociodemographic, housing variables, and the psychosocial variables of stress and social support. The
dependent variables were frequency and severity of
injuries. In this framework, endogenous and exogenous
determinants may be associated with injury outcome
through the moderating effect of parental safety practices. Specific hypotheses tested in this study were that
childhood injury risk was greater at: (a) higher level of
maternal stress and lower level of social support, (b)
lower level of Perceptions of Risk of Injury and lower
level of Perceptions of Risk of Hazard, and (c) lower level
of maternal supervision and higher number of homecontrollable hazards (lack of safety-proofing behaviors).
Method
This study utilized both secondary and primary sources
of data and was part of a larger population-based study
(Child Safety Study) that used a cross-sectional, randomdigit dial telephone survey in a medium-sized city in the
southeastern part of the United States between March
1988 and March 1989. Columbia, South Carolina, is
typical of many southeastern metropolitan areas today
whose populations are more educated and affluent than
those in adjacent rural areas, making them comparable
to other urban and suburban areas in the United States
with large middle-class populations.
Participants
Respondents were mothers of young children who had
participated in both the telephone survey (Child Safety
Study) and as part of an in-home subsample of mothers
(Home Safety Survey) who kept diaries about their
children’s injury experience. The Home Safety Study
consisted of a 25% randomly selected subsample of
families from the Child Safety Study; 75% of the mothers
in the randomly selected subsample agreed to participate, resulting in a total of 230 families. More details
about the Child Safety Study and the Home Safety Study
are described elsewhere (Glik et al., 1992; Glik et al.,
1991; Greaves et al., 1994). Table I presents a summary
of data sources and measures used.
Eligibility requirements for the Child Safety Study
families were as follows: residence in the two counties of
Childhood Unintentional Injuries: Factors
Table I.
Variables, Data Sources, and Data Collection Methods
Variable
Sociodemographic:
Income, race, maternal age, maternal employment,
marital status, and number of children
Psychosocial:
Study
Source
Child Safety Study
Reported data collected during telephone interviews
(RDD Survey)
(n = 1,247)
Child Safety Study
Reported data collected during telephone interviews
Child Safety Study
Reported data collected during telephone interviews
Child Safety Study
Reported data collected during telephone interviews
Child Safety Study
Reported data collected during telephone interviews
Home Safety Study
Observational data collected during home interviews
Maternal stress
Maternal coping
Maternal social support
Cognitive:
Perceptions of risk of hazard
Perceptions of risk of injury
Child characteristics:
Age, sex, child behavioral characteristics, number of
children in household, and type of child care
Housing:
Housing stock
Home repair
Safety Behaviors:
Home controllable hazards
(Home interviews) (n = 230)
Maternal supervision
Injuries:
Diary Sample (n = 159)
Diary data recorded by mothers
Data Collection
to record the dates of each injury on the front of
a monthly calendar and then write a brief description on
the back describing each injury: the site, the cause, the
severity, and the type of treatment required. Of the 230
mothers interviewed, 195 mothers completed and
returned completed diaries. For the purpose of this
study, only diaries that reported information between
the period September 1988 and August 1989 were
included in the analysis. An entire year was chosen to
account for seasonal variation in type of injuries. A total
of 159 diaries were available for analyses.
The data collection strategy used in this study was to
randomly choose one preschool child in each interviewed household to be the index child. Mothers were
asked to refer to this child when responding to the
telephone and home interviews and when completing
diaries for recording injuries.
It should be noted that this article represents
a further analysis of measures obtained in previous
studies that have not previously been reported in the
literature (see Glik et al., 1991; Glik et al., 1993; Greaves
et al., 1994). Unlike previous studies that focused on
parental perceptions and behaviors, this investigation
focused on injury events among young children.
During the home interviews, mothers in the subsample were asked to keep prospective diary data on
injuries suffered by their children. Mothers were asked
Measures
Measures used can be grouped into six sets: sociodemographic, housing related, psychosocial, cognitive, child
related, behavioral, and unintentional injuries.
Sociodemographic Variables. Social and demographic
characteristics of the sample are presented in Table II.
Housing Variables. These were observational measures
taken by home interviewers in the Child Safety Study.
The housing stock was a categorical variable that
indicated whether the home was a single-family dwelling, a trailer, or an apartment. Housing repair was
a dichotomous variable that related to whether the home
needed major structural repairs or not.
Psychosocial Variables. The 7-item Perceived Maternal
Stress scale examined stressors reported by mothers
the Columbia metropolitan area, having at least one
child between the ages of 6 months and 5 years, and an
operating telephone. Of the 1,475 eligible households,
1,247 agreed to participate, a response rate of 85%. The
Institutional Review Board of the University of South
Carolina approved the study, and full-informed consent
was obtained prior to data collection. Respondents were
restricted to mothers, to enhance comparability among
informants.
275
276
Dal Santo, Goodman, Glik, and Jackson
Table II. Social and Demographic Characteristics of Subjects
in the Study Sample (n = 159)
Variable
Frequency (n)
Percent (%)
Household annual income
,$15,000
19
11.9
$15,000–49,999
87
54.8
$50,000
53
33.3
Race
White
Nonwhite
121
76.1
38
23.9
137
86.2
22
13.8
Marital status
Married or living with significant other
Not married
Mother’s education
High school or less
40
25.2
Some college/college graduate
90
56.6
Graduate school
29
18.2
17–26
30
18.9
27–36
101
63.5
37–44
28
17.6
Employed outside the home
95
59.5
Not employed
64
40.5
1
53
33.3
2
70
44.0
3–5
36
22.7
Mother’s age
Mother’s employment
No. of children in the home
Child’s gender
Male
80
50.3
Female
79
49.7
Child’s age
6 months to less than 1 year
16
10.1
One year to less than 2.5 years
42
26.4
101
63.5
2.5 years to 5 years
House condition
Needs repair
Does not need repair
10
6.3
149
93.7
Type of child care
Stays at home with parent
69
43.4
Stays at home with an adult
25
15.7
Day care program
57
35.9
8
5.0
Other
(a 5 0.49), whereas the 7-item Perceived Maternal
Coping scale measured the degree to which the mother
felt she successfully coped during the past month (a 5
.70). These two scales are subscales, derived by factor
analysis of the 14-item Perceived Stress scale, a standardized measure with an alpha coefficient of 0.70, designed
by Cohen, Karmack, and Marmelstein (1983). The
Maternal Social Support Index measured perceived
social support from family and friends and had an alpha
coefficient of 0.53, compared to 0.63 reported by the
original author (Pascoe, Loda, Jeffries & Earp, 1987).
Cognitive Variables. Scales for measuring perceptions
of risk were developed by Glik et al. (1991) and were
conceptualized as having two domains: perceptions of
risk of injury and perceptions of risk of hazards. Each of
the two domains consisted of two Likert-scaled dimensions: chance (likelihood) and seriousness of injuries,
and likelihood and dangerousness (seriousness) of
hazards (Glik et al., 1991).
The Perceptions of Risk of Injury scale had 17
questions that asked mothers about injuries for preschool children, whereas the Perceptions of Risk of
Hazard scale had 19 questions that asked about hazards
associated with injuries for preschool children. The
Perceptions of Risk of Injury scale was computed by
multiplying each item in the Likelihood of Injury scale
by its counterpart in the Seriousness of Injury scale (e.g.,
perceived likelihood of a burn times perceived seriousness of a burn). Likewise, the Perceptions of Risk of
Hazard scale was developed by multiplying each item in
the Likelihood of Hazard scale by its counterpart on the
Dangerousness of Hazard scale (e.g., perceived likelihood of electrical outlets times perceived dangerousness
of electrical outlets). The rationale for this procedure
was that the risk of an injury or hazard has different
profiles of likelihood and seriousness—for example, a fall
may be likely but not serious, whereas injury from an
electrical outlet may be serious but not likely. Risk
perceptions of injuries and hazards also demonstrated
sufficiently high interitem correlation to be treated as
summed measures. The Perceived Risk of Injury scale
had an alpha coefficient of 0.86, whereas the Perceived
Risks of Hazard scale had an alpha coefficient of 0.89.
Child-related Variables. Of the child’s characteristics,
child’s age was recorded in 6-month increments ranging
from 6 months to 5 years. For multivariate analyses, this
variable was dichotomized as under 2.5 years and over 2.5
years. The child’s sex was a dichotomous variable. The
child’s behavioral characteristics as perceived by the
parents were measured using a 9-item semantic differential scale developed by Glik et al. (1991). The alpha
reliability of the scale was 0.63, with high scores
indicating that parents perceived their children to be
more active, more likely to take risks, and harder to
Childhood Unintentional Injuries: Factors
manage. Child’s injury history was measured by one
item on the Child Safety Questionnaire, which asked the
mother whether the child has had an injury serious
enough to be medically treated over the previous year.
The child’s day care variable had five categories and for
multivariate analyses was collapsed into two categories:
stay at home and part-time or full-time day care.
Behavioral Variables. Safety behaviors were measured
during home observations by trained interviewers in the
Home Safety Study. During the initial home visits, each
interviewer was accompanied by the investigator, and
training was continued until there was an 85% agreement
between the interviewer and the investigator on the
safety-hazard observations and the supervision scenarios.
Safety behaviors had two dimensions: home controllable hazards and self-reported supervision practices.
Home controllable hazards are hazards that are under
control of the occupants, such as keeping cleaning
supplies out of reach. The Home Controllable Hazards
scale was derived from an observational checklist adapted
from the Home Accident Prevention Inventory (HAPI)
developed by Tertinger, Green, and Lutzker (1984). This
measure is a proxy measure for lack of safety-proofing
behaviors. Some such hazards include uncovered electrical outlets, accessible matches or lighters, accessible
soaps and razors. A higher score on this scale indicates
more hazards. The alpha reliability coefficient for this
subscale was 0.65, which is somewhat lower than the
alpha coefficient of 0.77 reported in the Home Safety
Study. Self-reported supervision practices were measured
by a 10-item inventory of potentially hazardous situations
that elicited open-ended responses by mothers (Glik et
al., 1993). The 10 situations were chosen to represent
common situations that required a decision to act by the
mother to protect her child. The alpha coefficient of
internal consistency reliability for this instrument was
0.57. This instrument was reported to have an association
with control over household hazards (Glik et al., 1993).
Unintentional Injuries. The outcome measure was the
unintentional minor injury occurring to children under 5
years of age. An injury was defined as a discrete event that
produced pain lasting at least 10 minutes or discernable
tissue damage, such as a scratch or a bruise (Peterson,
Saldana, & Heilblum, 1996). Minor-injury data in this
study were classified by severity ratings of injuries into
two major categories: trivial injuries and serious injuries.
Severity ratings were determined using the Minor Injury
Severity scale for minor injuries, a 6-point rating scale
developed by Peterson et al. (1996). Severity ratings were
specific to type of injury and to the part of the body
injured, with high scores indicating more tissue damage.
Trivial injuries were defined as minor injuries that had
severity levels of 1 and 2. Serious injuries were defined as
those minor injuries that had severity levels 3 and higher.
The dependent variable was injury-free time (survival
time) or the time until the first serious injury occurred.
The measure of effect is the Hazard Ratio, used to
compare the relative hazard for individuals, with
different values of the independent variables.
Interrater reliability was performed for coding the
total number of injuries per household and for coding
the severity level of each injury. To measure intercoder
reliability, 20% of the diaries were transcribed and
recoded by another coder. There was a correlation of
0.96 between the two ratings in relation to the total
number of injuries per preschooler. The coding for
severity level within type of injury had a reliability of
Pearson r 5 0.83 across coders. This intercoder
reliability related to the ratings of injuries as trivial
versus serious. The intercoder reliability of the injury
ratings of this scale was reported as Pearson r 5 0.71
Data Analyses
Descriptive statistics included percentages, frequencies,
means, and standard deviations. Correlation coefficients
were calculated for all of the variables, to test for
multicollinearity. Survival analysis was used to measure
the effect of the explanatory variables on the injury
outcome. More specifically, this technique was used to
measure the effect of explanatory variables on survival
time (injury-free time). Survival analysis was chosen
because the usual parametric regression methods were
rendered inappropriate as a result of the variation in the
number of months of injury reporting (in diaries) and
the existence of censoring. Survival times were measured
as time, in months, to a serious injury event (severity
level 3). Injury-free time, for uncensored data, was the
period in months between first starting reporting and the
first serious injury event (injury of severity level 3).
Censored data contain households that reported no
serious injuries by the end of the study or who withdrew
or stopped reporting before the end of the 12-month
study period. For censored households, injury-free time
is equal to the period of reporting. Households were
eliminated from the analysis after the first serious event.
This eliminated the bias due to the dependence between
repeated events in the same household.
Both univariate and multivariate survival methods
were used. The associations between predictor variables
and the outcome were first examined individually using
277
278
Dal Santo, Goodman, Glik, and Jackson
univariate Cox proportional hazards models. The results
from the univariate analyses were then used with
a modified backward-elimination procedure to develop
multivariate models for estimating the multivariate
associations of the predictor variables and injury-free
time. The Product-Limit survival curves and log
( log[survival]) plots were also used to verify the
assumption of proportionality for the predictor variables. Since the Cox model is a model for the hazard of an
event, the measure of association is the hazard ratio,
which can be interpreted similarly to the relative risk.
Results
Social and demographic characteristics of the sample are
described in Table II. The families in this sample were
more affluent than the general population in South
Carolina, with 60% of the families having incomes above
$30,000. The sample had an even distribution of male
(50%) and female (50%) children. The total number of
injury events was 1,273, with 23 injury events that resulted in multiple injuries. The total number of reported
injuries was 1,299: 35% of the households reported no
injuries (n 5 12) or only trivial injuries (n 5 44) and were
thus considered censored households (n 5 56); 65% of the
households (n 5 103) reported one or more serious
injuries and were thus considered uncensored households.
The most frequent types of reported injuries were as
follows, in descending order: bruises (38.6%), scrapes
(32.3%), cuts (10.2%), and crushing injuries (4.3%).
Home injuries constituted 56% of total injuries: 36% were
home-indoor injuries and 20% were home-outdoor.
Almost half of the injuries (49%) involved falls. In sum,
3.5% of the injuries received medical treatment, 25% were
treated at home, and 70% were untreated.
Results of the univariate survival regression analysis
applying the Cox model showed that race, age of the
child, and maternal employment were the independent
variables that had a significant predictive effect on injuryfree time. Results were identical to those obtained using
the product limit method. These models indicated that
there was a significant difference in injury-free time by
race category; the median injury-free time was greater for
nonwhite preschoolers. Children above 2.5 years of age
had lower injury-free times than children below 2.5 years
of age. Preschoolers whose mothers worked full-time had
higher-median injury-free time than preschoolers whose
mothers worked part-time or did not work.
In the Cox multivariate model building, sociodemographic variables, perceptions of risk, maternal stress
and coping, social support, parental safety behaviors,
and child-related variables were entered in the initial
model. Six interaction terms that had to be fitted into the
initial model were created by multiplying each of the two
domains of the perceptions of risk with maternal
supervision, child injury history, and home controllable
hazards. In preliminary analyses interaction terms for
race, income, age of the mother, age of the child, and
child’s injury history were computed with maternal
perceptions of risk and safety behaviors and entered as
sets into regression equations: No interaction terms were
found to be significant.
Table III presents the results of the multivariate
regression analysis applying the Cox model. The model
coefficients, hazard ratios, and confidence intervals are
provided for all of the variables in the model. Of the
sociodemographic variables, only race, maternal employment, and home repair were significant predictors of
the hazard (injury risk). Results indicate that the hazard
(risk) of injury per unit time for white preschoolers was
almost three times that of nonwhite preschoolers. The
nonwhite preschoolers were predominantly African
Americans (83%). Children of mothers who were
unemployed or had part-time employment had
2.14 (1.278, 3.582) times the risk of injury per unit
time than children of mothers who were employed fulltime. Preschoolers whose homes needed repair had an
estimated risk of injury 3.92 (1.290, 11.953) times the
risk of injury of preschoolers whose homes did not need
repair. Maternal age, maternal education, number of
children, and marital status were not significant
predictors.
Age of the child and child behavioral characteristics
were significant predictors of injury risk. Preschoolers
over 2.5 years were 2.6 (1.448, 4.473) times as likely to
suffer a serious injury per unit time than younger
children, and children with difficult or hard-to-manage
behavioral characteristics were at higher risk of injury
than other children. Since child behavioral characteristics constituted a continuous variable, the hazard ratio
of 1.09 (1.044, 1.142) indicates that a preschooler with
more difficult behavioral characteristics (one unit higher
on the scale that measured child behavioral characteristics) had 1.09 times the risk of injury of another child
with less difficult behavioral characteristics when adjusting for other variables. For a difference of 10 units on the
child behavioral characteristics scale, the hazard ratio
would be 2.42.
Maternal Stress and Social Support. The psychosocial
variables of maternal stress, maternal coping, and
Childhood Unintentional Injuries: Factors
Table III.
Results of Multivariate Analysis for Predicting Injury-Free Time in Months
Variable
B
SE B
Hazard Ratioa
95% CI
Sociodemographic variables
Race (white)
1.028
0.358
2.796*
1.385, 5.641
0.761
0.263
2.141*
1.278, 3.582
1.368
0.568
3.922*
1.290, 11.953
Maternal employment
Not full-time
Home repair
Need for home repair
Child characteristic variables
Age . 2.5 years
0.934
0.288
2.545*
1.448, 4.473
Child behavioral characteristics
0.088
0.023
1.092**
1.044, 1.092
Perceptions of risk of hazard (PH)
0.021
0.034
1.021
0.955, 1.092
Perceptions of risk of injury (PI)
0.92
0.037
0.912*
0.370, 0.981
Cognitive variables
Behavioral variables
Maternal supervision (MS)
0.192
0.095
0.825*
0.685, 0.994
Home controllable hazards (HAZ)
0.385
0.114
0.680*
0.544, 0.850
PH 3 HAZ
0.002
0.001
1.002*
1.000, 1.004
PH 3 MS
0.001
0.001
0.999
0.999, 1.002
PI 3 MS
0.003
0.001
1.003*
1.001, 1.005
Interactions
Note. B = unstandardized regression coefficient.
a
For completeness, hazard ratios are given here for variables involved in interactions. These are not directly interpretable without considering multiple variables
simultaneously, as in Table IV.
* p , .05. ** p , .001.
maternal social support were not significant predictors
in the final Cox model. The low reliability of the
Perceived Maternal Stress scale (alpha of 0.49) may have
contributed to its lack of association, but maternal
coping with a higher alpha (0.70) was also not
significantly associated with injury in either univariate
or multivariate models.
Perceptions of Risk. Table III shows significant interactions between cognitive and behavioral variables. Thus,
these variables must be examined simultaneously. Table
IV is designed to aid in this interpretation. We can see
from Table IV that at low home controllable hazards (2)
and with maternal supervision fixed at 35, higher
perception of risk of hazard (a 50-unit increase) is
protective, reducing risk almost in half (hazard ratio 5
0.52, p , .01). At high home controllable hazards (10),
a similar increase in perceptions of risk of hazard has little
effect, and that effect was in the direction of increased risk
(relative hazard 5 1.09, p ,.01). Similarly, at low
maternal supervision, higher perception of risk of injury
(a 50-unit increase) is protective (relative hazard 5 0.51,
p , .01); but at high maternal supervision, increased
perceptions of risk of injury is associated with increased
risk (relative hazard 5 1.86, p , .01).
Maternal Safety Behaviors. Results of multivariate
analyses reveal that the effect of maternal supervision
on injury risk is moderated by perceptions of risk. Due to
interactions between cognitive and behavioral variables,
the hypothesis that the higher the maternal supervision,
the lower the injury risk is supported only at low levels
of perceptions of risk of injury. We can see from Table IV
that at low perceptions of risk of injury (60) and with
perceptions of risk of hazard fixed at 200, higher
maternal supervision (a 1-unit change) is protective
(relative hazard 5 0.78, p , .01). However, at high
perceptions of risk of injury (280), a 1-unit change in
maternal supervision is associated with increased risk
(relative hazard 5 1.38, p , .01). Also, due to interaction effects, an increase in home controllable hazards
is significantly related to higher injury risk only at high
levels of perceptions of risk of hazard. Table IV shows
that at high perceptions of risk of hazard (280), higher
home controllable hazards (a 1-unit change) is associated with increased injury risk (relative hazard 5 1.18).
279
280
Dal Santo, Goodman, Glik, and Jackson
Table IV. Selected Point Estimates to Demonstrate the Effects
of Significant Interaction Terms
Variable
Hazard Ratio
Perceptions of risk of hazard (50-unit change)
p
0.002
(Maternal supervision fixed at 35)
At low home controllable hazards (2)
0.52
At high home controllable hazards (10)
1.09
Perceptions of risk of injury (50-unit change)
0.008
At low maternal supervision (30)
0.51
At high maternal supervision (40)
Maternal supervision (1-unit change)
1.86
0.008
(Perceptions of risk of hazard fixed at 200)
At low perceptions of risk of injury (60)
0.78
At high perceptions of risk of injury (280)
1.38
Home controllable hazards (1-unit change)
0.002
At low perceptions of risk of hazard (60)
0.82
At high perceptions of risk of hazard (280)
1.18
However, at low perceptions of risk of hazard (60),
a similar increase in home controllable hazards is protective (relative hazard 5 0.82, p , .01).
Discussion
Findings reveal that childhood injury is a complex
phenomenon predicted by a set of sociodemographic,
cognitive, behavioral and child-related factors. Both
endogenous (cognitive and behavioral) variables and
exogenous determinants (sociodemographic and housing variables) in the injury prevention conceptual
model, on which this study was based, are associated
with childhood injuries.
Of the sociodemographic characteristics, race,
maternal employment, and home repair were significantly associated with injury risk. The findings that
white children were almost at three times the risk of
injury than nonwhite children are in contrast with those
related to fatal injuries, which are higher among
nonwhite preschoolers (Kennedy & Lipsitt, 1998;
Morrongiello & Rennie, 1998; Rodriguez & Brown,
1990). However, our findings are consistent with the
findings of a national study (N 5 11,630) on unintentional injuries based on data from the 1988
National Health Interview Survey (Bussing, Menvielle,
& Zima, 1996). In the national study, rates of injuries
that required medical treatment were higher in white
children (17.9%) than among African American (9.3%)
or Hispanic (9.3%) children. The differences among
injury rates may be attributed to true rate differences, or
they may be the result of bias, including differential
reporting of injuries (i.e., a different threshold for injury
reporting) between white and African American mothers
(Bussing et al., 1996). The finding that preschoolers of
employed mothers who worked full-time were at lower
injury risk than other preschoolers may be explained by
the fact that the majority of injuries to children take
place at home rather than at day care (Leland, Gerrard,
& Smith, 1993). However, the possibility of employed
mothers’ underreporting daycare injuries cannot be
ruled out. Employed mothers either did not have
firsthand knowledge of minor injuries suffered by their
children or simply did not record the incidents in the
diaries because of inadequate information.
The finding that preschoolers whose homes needed
repair were at 3.8 times the risk of injury than
preschoolers whose homes did not need repair points
out the salient role of the physical environment in
predicting childhood injuries. This predictive role is well
documented (Rivara, 1982). One other interesting
finding is that the child’s age and behavioral characteristics were significantly associated with increased injury
risk. The finding that preschoolers older than 2.5 years
were at higher risk of injury than younger ones is
consistent with other studies indicating that children’s
injury risk increases sharply during the first 2 years with
a peak occurring between 2 and 4 years of age (Matheny,
1987). As the child gets older, exposure to risk increases
because in the process of development, the child’s
physical growth and increased motor skills are not
matched by an equivalent increase in the child’s
cognitive skills and appreciation of hazards (Grossman
& Rivara, 1992). In this study, age of the child over 2.5
years was negatively associated with maternal supervision (r 5 0.27, p , .01) and positively associated with
home controllable hazards (r 5 0.22, p , .01). One can
infer that as children grow older, mothers relax their
safety standards as well as their supervisory styles
because they overestimate their children’s cognitive
capabilities.
The significant association of child behavioral
characteristics with injury risk supports recommendations by previous researchers that child behavioral
characteristics be considered as a potential predictor in
childhood injury research (Bussing et al., 1996;
Matheny, 1987). Other studies report that overactivity
and aggressive behavior increased the risk of childhood
injuries (Grossman & Rivara, 1992). Although child
characteristics cannot be changed, the link between age,
sex, and behavioral characteristics of the child and injury
risk suggests the need for including child development
Childhood Unintentional Injuries: Factors
and behavioral management strategies in child safety
interventions.
An intriguing aspect of these findings is that there
were significant interactions between maternal perceptions of risk and safety behaviors when predicting injury
risk. As a consequence, our findings partially support the
hypotheses that increased perceptions of risk lowered
childhood injury risk and that increased safety behaviors
decreased injury risk among preschoolers. Results
revealed the importance of maternal supervision as
a moderator between maternal perceptions of risk of
injury and childhood injury risk. Increased perceptions
of risk of injury resulted in lower injury risk only at low
levels of maternal supervision. It is notable, however,
that this relationship did not hold at high levels of
maternal supervision.
Similarly, findings showed that home controllable
hazards moderated the effect of perceptions of risk of
hazard on injury risk. Perceptions of risk of hazard were
associated with less injury risk only at a low level of
home controllable hazards. At higher levels of home
controllable hazards, increased perceptions of risk of
hazard were associated with higher injury risk. This may
reflect a lack of motivation or means to reduce high
levels of home hazards (i.e., to engage in safety-proofing
behaviors), even when the mothers were aware of these
hazards and perceived their children to be at high risk of
such hazards. An equivalent interpretation is that the
hypothesis that the higher the home controllable
hazards, the higher the injury risk is true only at high
levels of perceptions of risk of hazard.
Findings concerning maternal supervision indicate
that perceptions of risk of injury moderated the effect of
maternal supervision on injury risk. The hypothesis that
the higher the maternal supervision, the lower the injury
risk is supported only at low levels of perceptions of risk
of injury. One possible explanation is that if perceptions
of risk of injury are considered a function of the maternal
perceptions of how risky the environment the child is in,
then maternal supervision is beneficial in reducing
injuries when perceptions of risk of injury are low
(e.g., falling against furniture). However, when maternal
perceptions of risk of injury are high (e.g., falling from
a bike), a higher level of maternal supervision does not
reduce injury risk. This implies that the beneficial effect
of maternal supervision in reducing injury risk is
protective in environments perceived by mothers to be
of low risk of childhood injury. Other studies pointed
out that parental supervision is not preventive for all
injuries. Parents, medical personnel, and child pro-
tection service workers do consider that supervision
level is dependent on age of the child and on how risky
the environment is (Peterson, Ewigman, & Kivlahan,
1993).
These findings pose a challenging task for future
researchers to explicate the relationship between cognitive and behavioral predictors and childhood injuries
using path analysis or structural equations or models.
The Cox proportional hazard model and extensions of
these models to time-dependent variables and to repeated events should form an important tool in further
analyses of this type, as such methods require knowing
the time of the event rather than just whether the event
occurs or not. The interactions between parental
cognitive and behavioral variables suggest the need for
pediatric psychologists to consider interactional models
that explain the mechanisms by which cognitive,
behavioral, and situational risk factors interact to put
children at risk of unintentional injuries. One other
challenging area for future research is to test whether
this predictive model is applicable to fatal and near-fatal
childhood injuries and to study the link between minor
childhood injuries and serious injuries.
Some of the study limitations are the small sample
size of the diary study and the underrepresentation of
low-income households in the study because households
with no telephones were not included in the telephone
survey. Also, since the home safety sample was slightly
biased with regards to mothers’ educational level, the
diary sample also included women that had slightly
higher education than the general population sample.
Another limitation is the focus of the study solely on
mothers as respondents—ignoring fathers, stepfathers,
care providers, and other significant figures. The age of
the data is a limitation, but findings are still significant as
behavioral research reveals that parental perceptions of
risk of childhood injury and parental safety behaviors
have been underinvestigated (Peterson, DiLillo, Lewis,
& Sher, 2002; Peterson & Stern, 1997). There is paucity
of research on parental practices and child injury
prevention at the family level (Peterson & Stern,
1997). Furthermore, there is scarcity of data linking
parental cognitions and behaviors to actual child injury
outcome. This paper attempts to address this gap in
knowledge. Of particular concern is the low reliability of
some of the measures—specifically, these for perceived
maternal stress, home controllable hazards, and perceived child behavioral characteristics.
Even though our conclusions are tentative—due to
certain measurement deficiencies such as the low in-
281
282
Dal Santo, Goodman, Glik, and Jackson
ternal consistency reliability of some of the measures—
our data suggest that there is a parental role in injury
control, a role that still needs to be more clearly defined.
Our findings support the notion that child safety
interventions should be comprehensive, considering
both the behavioral approach and the active involvement
of parents and caregivers as well as passive environmental and legislative approaches.
Author’s Note
Janet Abboud Dal Santo, Department of Health Promotion and Education, School of Public Health,
University of South Carolina; Robert M. Goodman,
Department of Health Promotion and Education, School
of Public Health, University of South Carolina; Deborah
Glik, Department of Community Health Sciences,
School of Public Health, University of California at Los
Angeles; Kirby Jackson, Department of Epidemiology
and Biostatistics, School of Public Health, University of
South Carolina.
Janet Abboud Dal Santo is now at the Injury
Prevention Research Center, University of North Carolina at Chapel Hill. Robert Goodman is now at the
Department of Behavior and Community Health, Graduate School of Public Health, University of Pittsburgh.
Acknowledgments
This research was partially supported by U.S. Public
Health Service Grant R01-HD 21599-03, from the National Institute of Child Health and Human Development.
We would like to thank Michael Bowling at the
School of Public Health, University of North Carolina at
Chapel Hill, and Beth Morocco of the Pacific Institute for
Research and Evaluation, for their review and critique of
earlier versions of this manuscript. Michael Bowling
provided helpful comments with regard to the discussion of results in this paper.
Received July 1, 2002; revisions received November 5, 2002,
March 4, 2003, and July 3, 2003; accepted July 9, 2003
References
Bussing, R., Menvielle, E., & Zima, B. (1996).
Relationship between behavioral problems and
unintentional injuries in US children. Archives of
Pediatrics and Adolescent Medicine, 150, 50–56.
Centers for Disease Control and Prevention,
National Center for Injury Prevention and Control.
(2001). Injury fact book 2001–2002. Retrieved March
17, 2004, from http://www.cdc.gov/ncipc/
fact_book/.
Cohen, S., Karmack, T., & Marmelstein, N. (1983). A
measure of perceived stress. Journal of Health and
Social Behavior, 23(4), 385–396.
Fonseca, S., Victora, C., Halpern, R., Lima, R., &
Barros, F. (2002). Comparison of two methods
for assessing injuries among pre-school children.
Injury Prevention, 8, 79–82.
Garling, A., & Garling, T. (1993). Mother’s supervision
and perception of young children’s risk of
unintentional injury in the home. Journal of Pediatric
Psychology, 18(1), 105–113.
Glik, D., Greaves, P., Kronenfeld, J., & Jackson, K.
(1992). Safety hazards in households with young
children. Journal of Pediatric Psychology, 18(1),
115–131.
Glik, D., Kronenfeld, J., & Jackson, K. (1991). Predictors
of risk perceptions of childhood injury among
parents of preschoolers. Health Education Quarterly,
18(3), 285–301.
Glik, D., Kronenfeld, J., & Jackson, K. (1993). Safety
behaviors among parents of preschoolers. Health
Values, 17(1), 18–26.
Greaves, P., Glik, D., Kronenfeld, J., & Jackson, K.
(1994). Determinants of controllable in-home
child safety hazards. Health Education Research,
9(3), 307–314.
Grossman, D. (2000). The history of injury control
and the epidemiology of child and adolescent
injuries. In R. Behrman (Ed.), The future of children:
Vol. 10, no. 1. Unintentional injuries in childhood
(pp. 23–52). Los Altos, CA: David and Lucile
Packard Foundation.
Grossman, D., & Rivara, F. (1992). Injury control in
childhood. Pediatric Clinics of North America,
39(3), 471–485.
Harel, Y., Overpeck, M. D., Jones, D., Scheidt, P.,
Bijur, P., Trumble, A., & Anderson, J. (1994).
The effect of recall bias on estimates of annual
non-fatal injury rates for children and
adolescents. American Journal of Public Health,
84, 599–605.
Janz, N., & Becker, M. (1984). The Health Belief
Model: A decade later. Health Education Quarterly,
11(1), 1–47.
Kennedy, C., & Lipsitt, L. (1998). Risk taking in
preschool children. Journal of Pediatric Nursing,
13(2), 77–84.
Childhood Unintentional Injuries: Factors
Kronenfeld, J., & Glik, D. (1990). Perceptions of risk:
Its applicability in medical sociological research.
In D. Wertz, Advances in the sociology of health
care systems: Vol. 9 (pp. 303–330). Greenwich,
CN: JAI Press.
Leland, N., Garrard, J., & Smith, D. (1993). Injuries
to preschool-age children in day-care centers.
American Journal of Diseases of Children, 147,
826–831.
Marsh, P., & Kendrick, D. (1999). Using a diary to
record near misses and minor injuries—which
method of administration is best? Injury Prevention,
5, 305–309.
Matheny, A. P. (1986). Injuries among toddlers:
Contributions from child, mother, and family.
Journal of Pediatric Psychology, 11, 163–176.
Matheny, A. P. (1987). Psychological characteristics of
childhood accidents. Journal of Social Issues,
43(2), 45–60.
Morrongiello, B., & Rennie, H. (1998). Why do boys
engage in more risk taking than girls? The role of
attributions, beliefs and risk appraisals. Journal of
Pediatric Psychology, 23(1), 33–43.
National Safety Council. (2001). Injury facts (NSC
Press Product No. 02301-0000). Itasca, IL:
Author.
Pascoe, J., Loda, F., Jeffries, V., & Earp, J. (1987). The
association between mothers’ social support and
provision of stimulation to their children.
Developmental and Behavioral Pediatrics, 8, 207–212.
Peterson, L., DiLillo, D., Lewis, T., & Sher, K. (2002).
Improvement in quantity and quality of prevention
measurement of toddler injuries and parental
interventions. Behavior Therapy, 33, 271–297.
Peterson, L., Ewigman, B., & Kivlahan, C. (1993).
Judgments regarding appropriate child supervision
to prevent injury: The role of environmental risk
and child age. Child Development, 64, 934–950.
Peterson, L., Saldana, L., & Heilblum, N. (1996).
Quantifying tissue damage from childhood injury:
The minor injury severity scale. Journal of Pediatric
Psychology, 21(2), 251–267.
Peterson, L., & Stern, B. (1997). Family processes and
child risk for injury. Behavior Research and Therapy,
35(3), 179–190.
Reading R., Langford, I., Haynes, R., & Lovett A. (1999).
Accidents in preschool children: Comparing family
and neighborhood risk factors. Social Science and
Medicine, 48(3), 321–330.
Rivara, F. (1982). Epidemiology of childhood injury.
American Journal of Diseases in Children, 136,
399–405.
Rodriguez, J., & Brown, S. (1990). Childhood injuries in
the United States. American Journal of Diseases in
Children, 144, 627–646.
Slovic, P., Fischhoff, B., & Lichtenstein, S. (1978).
Accident probabilities and seat belt usage: A
psychological perspective. Accident Analysis and
Prevention, 10, 281–285.
Slovic, P., Fischhoff, B., & Lichtenstein S. (1982). Why
study risk perception? Risk Analysis, 2(2), 83–93.
Tertinger, D., Green, B., & Lutzker, J. (1984). Home
safety: Development and validation of one
component of an ecobehavioral treatment program
for abused and neglected children. Journal of Applied
Behavior Analysis, 17(2), 159–174.
Verbrugge, L. (1980). Health diaries. Medical Care,
18, 73–94.
283