Measuring behaviours for escaping from house fires: use of latent variable models to summarise multiple behaviours
© Ploubidis et al. 2015
Received: 28 January 2015
Accepted: 30 November 2015
Published: 15 December 2015
This paper reports the development and testing of a construct measuring parental fire safety behaviours for planning escape from a house fire.
Latent variable modelling of data on parental-reported fire safety behaviours and plans for escaping from a house fire and multivariable logistic regression to quantify the association between groups defined by the latent variable modelling and parental-report of having a plan for escaping from a house fire. Data comes from 1112 participants in a cluster randomised controlled trial set in children’s centres in 4 study centres in the UK.
A two class model provided the best fit to the data, combining responses to five fire safety planning behaviours. The first group (‘more behaviours for escaping from a house fire’) comprised 86 % of participants who were most likely to have a torch, be aware of how their smoke alarm sounds, to have external door and window keys accessible, and exits clear. The second group (‘fewer behaviours for escaping from a house fire’) comprised 14 % of participants who were less likely to report these five behaviours. After adjusting for potential confounders, participants allocated to the ‘more behaviours for escaping from a house fire group were 2.5 times more likely to report having an escape plan (OR 2.48; 95 % CI 1.59–3.86) than those in the “fewer behaviours for escaping from a house fire” group.
Multiple fire safety behaviour questions can be combined into a single binary summary measure of fire safety behaviours for escaping from a house fire. Our findings will be useful to future studies wishing to use a single measure of fire safety planning behaviour as measures of outcome or exposure.
Trial registration number: NCT 01452191. Date of registration 13/10/2011
KeywordsFire escape plans Fire safety planning behaviours Latent variable models Methods
Globally fire-related burns resulted in around 96,000 deaths in children and young people aged under 20 years in 2004  and they are the 11th leading cause of death in children aged 1–9 years . While most fire-related deaths occur in low and middle-income countries , they remain an important public health problem in high income countries. House fires are an important public health problem in Great Britain, with more than 43,000 occurring in 2011/12, resulting in 287 deaths and 11,300 casualties across all ages . There steep social gradients in fire-related deaths in childhood; children of parents who have never worked or are long-term unemployed have fire-related death rates 38 times higher than those with parents in higher managerial or professional occupations . Data from the USA demonstrates smoke alarms are associated with a two to threefold lower risk of death in house fires [4, 5] and household smoke alarm ownership has increased in Britain from 8 % in 1988 to 86 % in 2008 . Given high levels of smoke alarm ownership, the emphasis of community fire safety is shifting towards the development of plans to help people escape safely from dwelling fires. In the UK, most Fire and Rescue Services offers free home fire safety checks, including help to develop a plan for escaping from a house fire . More than 770,000 homes in England received a home fire safety check in 2011–12 .
Few published studies have evaluated the impact of interventions promoting planning for escaping from a house fire. A recent systematic review found only four studies evaluating home safety interventions that reported having or practising a plan for escaping from a house fire as an outcome measure [8–11]. Meta-analysis of these studies found home safety education was effective in increasing the proportion of families with a plan for escaping from a house fire , but this proportion remained relatively low post intervention, ranging across studies from 30 %  to 63 % . Furthermore, none of the studies defined what a plan for escaping from a house fire consisted of, and all used single item questions to assess whether the family had, or had practised, a plan.
We are currently conducting a cluster randomised controlled trial evaluating a fire safety intervention delivered by children’s centres in England to families with a child aged under 3 years. The primary outcome measure for the trial is whether the family reports having a plan for escaping from a house fire. We also measure self-reported behaviours which may form elements of a plan for escaping from a house fire, such as knowing how the smoke alarm sounds, keeping a torch next to the bed, making sure exits are clear, and that the keys for the external doors and windows are readily available . This paper reports the development and testing of a construct measuring parental fire safety behaviours for planning escape from a house fire.
Home safety questionnaires often contain multiple questions about the same construct (e.g. asking families about owning a smoke alarm and about owning a fire extinguisher, two elements of the construct of fire safety). In situations such as this, statistical data reduction methods which allow multi-dimensional data sets to be made smaller, without any loss of information, may be useful. We used latent variable models as a data reduction technique to derive a single summary measure of behaviours for planning escape from a house fire.
Sample and measures
Characteristics of participants
Number of smokers in household
Number of adults in household
3 or more
Number of children in household
4 or more
Age of youngest child
Under 1 year
Mother aged 16–20
Heavy drinker in householda
English as 1st language
Frequency of fire safety behaviours
Knows sound of alarm
External door keys accessible
Window keys accessible
Child might hide under bed
Child might hide in cupboard/wardrobe
A latent variable model is any model that includes unobserved random variables. The latent variable is not part of the dataset, in that it is not observed, but it is specified in order to account for any associations between the observed variables. Latent variable models combine information from different observed variables without making assumptions about units of measurement, and they also allow assessment of reliability and validity of these variables [14, 15]. To an increasing extent, latent variable modelling has been recognized as a valuable tool in epidemiological research  and numerous studies have employed latent variables to reduce a large number of observations and derive meaningful summaries. Other areas of application include survival analysis, meta-analysis, disease mapping, biometrical genetics, covariate measurement error models and joint models for longitudinal change and dropout.
There are two broad categories of latent variable models: dimensional models, where the latent variable is continuous and individuals are ranked (also called factor analysis or item response models); and discrete models, where the latent variable is categorical and individuals are grouped (also called latent class models). Hybrid models where both continuous and discrete latent variables are used are also available but their use is beyond the scope of this paper.
We first used a dimensional model to create a continuous latent plan for escaping from a house fire summary measure based on a combination of binary and ordinal fire safety variables. These comprised responses from seven questions asking about fire safety relevant to planning escape from a house fire (Table 2). To find the fire safety variables most compatible to the data latent structure, we used a discrete latent variable model to derive a categorical summary measure of behaviour for planning escape from a house fire, where individuals are grouped. Due to the binary and ordinal nature of the fire safety behaviour variables, in both approaches two parameter binary and ordinal logistic link functions were used.
In the dimensional models, ‘factor loadings’ represent the strength of the association between the fire safety variable and the latent plan for escaping from a house fire summary measure (and may be interpreted as correlations). ‘Thresholds’ represent the level of the latent plan for escaping from a house fire summary measure that must be reached for a specific response (“three days a week” for example) in a categorical or ordinal fire safety variable to be given. Each individual is assigned a score on each dimension, these scores in theory range from −∞ to +∞, but in most applications the range is from −3 to 3.
In discrete models, “posterior or conditional item probabilities” represent the probability, conditional on group membership, of a category of an observed variable (having a torch for example) being given. Class probabilities specify the relative size of each class, or population prevalence of each group. Group membership is based on the observed response pattern of items under an important assumption, called the conditional or local independence assumption, which implies that the correlation among the observed variables is explained by the latent categorical variable .
There are a range of tests that can be used to assess how well the model fits the data with different tests for dimensional and latent class models. Statistics for these measures are given in Tables 5 and 6, along with the values which suggest a good fit of the models to the data for each approach. All models were estimated with the “complex survey design” option in the Mplus 7.0 software  which accommodates complex sampling designs and returns robust parameter estimates taking into account dependence due to clustering.
As an external validation criterion for our binary summary measure, we used a question from the same questionnaire in which participants were asked whether they had a plan for escaping from a house fire. Using a multivariable logistic regression model we estimated the association between parents reporting that they had a plan for escaping from a house fire and the binary summary measure, adjusting for a range of confounding variables (as listed in Table 4).
Ethical approval was provided by the NRES East Midlands Committee (Derby), reference number 11/EM/0011.
Trial registration number: NCT 01452191. Date of registration 13/10/2011
A total of 347 (32 %) of the participants reported that they had a torch beside the bed and 1006 (98 %) reported that they were familiar with the sound of their alarm. A total of 838 (79 %) participants reported that they kept their exit routes clear, 926 (86 %) that they had the external door keys and 796 (75 %) that they had window keys in accessible places on at least 6 days per week, 308 (36 %) strongly agreed or agreed that, in the event of a fire, their child might hide under the bed and 301 (35 %) that their child might hide in a cupboard or wardrobe (Table 2).
Having external door and window keys accessible and keeping exits clear were the strongest indicators of the first plan for escaping from a house fire dimension The second plan for escaping from a house fire dimension comprised only two (highly loading) variables related to the likelihood that children may hide under a bed, or in a cupboard, in the event of a fire. The first and second plan for escaping from a house fire dimensions were not correlated with each other, suggesting their component questions were measuring different constructs. As the second dimension included only two variables unrelated to the first dimension (which included variables most relevant to planning for escape from a house fire), we excluded the variables indicating whether a child might hide from further analysis.
Posterior probabilities derived from categorical latent variable model
Class 1—more behaviours for escaping from a house fire
Class 2—fewer behaviours for escaping from a house fire
Knows sound of alarm
External door keys accessible
Window keys accessible
By contrast, the second group (‘fewer behaviours for escaping from a house fire) comprised 13 % of participants who were less likely to have a torch, and less likely to be aware of how their smoke alarm sounds. They were also less likely to have door and window keys accessible, and less likely to have exits clear. A typical member of this group does not have a torch, is not aware of how their smoke alarm sounds, has accessible door and window keys less than 1 day a week, and has exits clear less than 1 day a week.
Odds ratios (with 95 % CIs) for reporting having a plan for escaping from a house fire according to whether allocated to the ‘more behaviours for escaping from a house fire’ group (adjusted for all confounders presented in the table)
Odds ratio (95 % CI)
More behaviours for escaping from a house fire
>2 smokers in household
>2 adults in household
>2 children in household
Youngest child 1–2 years old
Mother >20 years old
Heavy drinker in household
English not first language
Non-white ethnic group
Index of multiple deprivation score
We have shown that multiple questions asking about behaviours for escaping from a house fire can be combined into a single binary summary measure, which provides a good fit to the data from the individual questions and it has been externally validated by comparing with a self-reported measure of having a plan for escaping from a house fire. Our findings suggest questions on having a torch beside the bed, knowing the sound of the smoke alarm, keeping exits clear and keeping door and window keys accessible appear to assess component elements of a plan for escaping from a house fire. Our findings will be useful to future studies wishing to use a single measure of planning for escape from a house fire as measures of outcome or exposure.
Strengths and limitations
To our knowledge this is the first published paper describing the development of a measure of planning for escape from a house fire. Previous studies [8–11] have reported only single-item measures asking whether families had, or had practised a plan for escaping from a house fire. Our measure is more comprehensive than these as it uses information on range of behaviours relevant to escaping from a house fire. As we also asked families whether they had a plan for escaping from a house fire we were able to validate our newly developed measure against the reports of having a plan for escaping from a house fire, including adjusting for possible confounders. However, the reports of having an escape plan were collected on the same questionnaire as the individual items and are not therefore independent. To provide a better external validation, we would have collected reports of escape plans on a different occasion to when the questionnaire was administered.
Our multivariable model found that families with more than 2 children were less likely to report having a plan for escaping from a house fire. It is plausible that making a plan to escape from a house fire is more difficult in larger families, as there are more children (likely to be at different developmental stages) to plan for and to explain and practise the plan with. The purpose of this analysis was to provide some validation of our summary measure, not to explore factors associated with having a plan to escape a house fire. This finding should therefore be interpreted with caution and requires replication in other studies.
Our behaviours for planning escape from a house fire were self-reported, with most questions asking about the frequency of behaviours and one question asking about knowledge of how smoke alarms sound. Validating these, for example, by home observations, is not possible, and to our knowledge there are no published studies reporting reliability or validity of these measures [19–26]. Two studies have reported some measures of reliability and validity of self-reported plans for escaping from a house fires. The first found a high level of agreement (91 %; no kappa coefficients reported) to repeat administration (over 14–24 days) of a single item question (Do you have a plan for escape from the home in the event of a fire?) as part of a general home safety survey amongst 35 parents of children aged 1–4 years . This question is very similar to that asked in our trial, suggesting the test–retest reliability of our question about having a plan for escaping from a house fire may be reasonable. An Australian study reported no significant difference in reported prevalence of having a fire evacuation plan on a telephone survey and home observation, although they did not report the percentage agreement, kappa coefficients, sensitivity, specificity or predictive values . It is possible that parents in our study over-reported behaviours they perceived to be socially acceptable. This could affect our findings if over-reporting varied substantially between the safety behaviours comprising our summary binary measure, although there is no reason to believe some of these safety behaviours are much more socially acceptable than others.
Our study was restricted to families with young children attending children’s centres in England. Children’s centres provide integrated early years health, social care and education to families living in disadvantaged areas. So, whilst our findings are applicable to families whose children are at greater risk of fire-related injury, they may not be generalisable to families with older children, more advantaged populations, to those from low and middle-income countries, or to planning for escaping from a house fires amongst other vulnerable populations such as the elderly or disabled.
The distribution of the first plan for escaping from a house fire dimension was highly skewed and resembled a mixture of distributions rather than a continuum. Using this distribution even with appropriate link functions as an outcome in predictive models would have been potentially biased since the two suspected mixture components may be associated with different sets of predictors. We therefore decided to use a categorical latent variable model to create our single summary measure.
Implications for research and practice
Our findings will be useful for future studies in high income countries where planning for escape from a house fire will be similar to that in the UK and for populations comprising families with small children. Such studies can use our questions to develop a single summary measure of fire escape planning behaviour. Many injury prevention interventions are evaluated using tools which measure multiple safety behaviours. The methods we describe could be used to develop single outcome measures for evaluation studies, or single measures of exposures for observational studies. This may be helpful in minimising the risk of “spuriously” significant findings resulting from multiplicity (statistical significance testing of multiple outcomes or exposures). Our findings will be of use to injury prevention practitioners who help families to develop plans for escaping from a house fire, as our questions could be used in home fire safety risk assessments or home safety check lists, and could be used by practitioners to assess the impact of their work on planning escape from house fires.
Multiple fire safety behaviour questions can be combined into a single binary summary measure of planning for escape from a house fire. Our findings will be useful to future studies wishing to use a single measure of behaviours for planning for escape from a house fire as measures of outcome or exposure.
GBP undertook the data analysis, contributed to interpretation of the data and to drafting the paper. PE contributed to the analysis and interpretation of the data and to drafting the paper. DK had the original idea for the trial, designed the trial, supervised data collection, contributed to the analysis and interpretation of the data and to drafting the paper. All authors read and approved the final manuscript.
This paper presents independent research commissioned by the National Institute for Health Research (NIHR) under its Programme Grants for Applied Research funding scheme (RP-PG-0407-10231). The views expressed in this article are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
Keeping Children Safe Study Group Joanne Ablewhite1, Penny Benford1, Clare Bryan1, Rose Clacy1, Carol Coupland1, Nicola Cooper2, Toity Deave3, Trudy Goodenough3, Adrian Hawkins4, Mike Hayes5, Paul Hindmarch4, Bryony Kay6, Benita Laird-Hopkins3, Gosia Majsak-Newman7, Elaine McColl8, Lisa McDaid7, Richard Reading9, Jane Stewart1, Alex J Sutton2, Elizabeth Towner3, Michael Watson10, Persephone Wynn1
1Division of Primary Care, School of Medicine, Tower Building, University Park, Nottingham, NG7 2RD, UK
2Department of Health Sciences, University of Leicester, Leicester LE1 7RH, UK
3Centre for Child and Adolescent Health, Health and Life Sciences, University of the West of England, Bristol, Oakfield House, Oakfield Grove, Bristol BS8 2BN, UK
4Great North Children’s Hospital, Research Unit Level 2, New Victoria Wing, Royal Victoria Infirmary, Queen Victoria Road, Newcastle upon Tyne, NE1 4LP, UK
5Child Accident Prevention Trust, Canterbury Court (1.09), 1-3 Brixton Road, London, SW9 6DE, UK
6Emergency Department, Bristol Royal Hospital for Children, Paul O’Gorman Building, Upper Maudlin Street, Bristol, BS2 8BJ
7NHS Clinical Research and Trials Unit, Norwich Medical School, University of East Anglia, Norwich, NR4 7TJ, UK
8Newcastle Clinical Trials Unit, 4th Floor, William Leech Building, The Medical School, Newcastle University, Framlington Place, Newcastle upon Tyne, NE2 4HH, UK
9Jenny Lind Paediatric Department, Norfolk and Norwich University Hospital, Colney Lane, Norwich, NR4 7UY, UK
10School of Health Sciences, University of Nottingham, D86, Queen’s Medical Centre, Nottingham, NG7 2HA, UK
The authors declare that they have no competing interests.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Peden M, Oyegbite K, Ozanne-Smith J, Hyder A, Branche C, Rahman A, Rivara F, Bartolomeos K. World report on child injury prevention. In: World Health Organisation, UNICEF, Geneva. 2008.Google Scholar
- Department for Communities and Local Government: Fire Statistics 2011 to 2012. In: London. 2012.Google Scholar
- Edwards P, Roberts I, Green J, Lutchmun S. Deaths from injury in children and employment status in family: analysis of trends in class specific death rates. BMJ. 2006;333(7559):119–21.PubMed CentralView ArticlePubMedGoogle Scholar
- Runyan CW, Bangdiwala SI, Linzer MA, Sacks JJ, Butts J. Risk factors for fatal residential fires. N Engl J Med. 1992;327(12):859–63.View ArticlePubMedGoogle Scholar
- Marshall SW, Runyan CW, Bangdiwala SI, Linzer MA, Sacks JJ, Butts JD. Fatal residential fires: who dies and who survives? JAMA J Am Med Assoc. 1998;279(20):1633–7.View ArticleGoogle Scholar
- http://www.fireservice.co.uk/safety/hfsc. Accessed 21 Aug 2013.
- Department for Communities and Local Government. FRS operational statistics bulletin for England 2011-12. London: FRS; 2012.Google Scholar
- Campbell NR, Ayala GX, Litrownik AJ, Slymen DJ, Zavala F, Elder JP. Evaluation of a first aid and home safety program for hispanic migrant adolescents. Am J Prev Med. 2001;20(4):258–65.View ArticlePubMedGoogle Scholar
- Hwang V, Duchossois G, Garcia-Espana J, Durbin D. Impact of a community based fire prevention intervention on fire safety knowledge and behavior in elementary school children. Inj Prev. 2006;12(5):344–6.PubMed CentralView ArticlePubMedGoogle Scholar
- Petridou E, Tolma E, Dessypris N, Trichopoulos D. A controlled evaluation of a community injury prevention project in two Greek islands. Int J Epidemiol. 1997;26(1):173–9.View ArticlePubMedGoogle Scholar
- Posner JC, Hawkins LA, Garcia-Espana F, Durbin DR. A randomised, clinical trial of a home safety intervention based in an Emergency Department setting. Pediatrics. 2004;113(6):1603–8.View ArticlePubMedGoogle Scholar
- Kendrick D, Young B, Mason-Jones AJ, Ilyas N, Achana FA, Cooper NJ, Hubbard SJ, Sutton AJ, Smith S, Wynn P et al: Home safety education and provision of safety equipment for injury prevention. Cochrane Database Syst Rev. 2012(9).Google Scholar
- Communities and Local Government: http://www.direct.gov.uk/en/groups/dg_digitalassets/@dg/@en/documents/digitalasset/dg_073993.pdf. Accessed 25 Aug 2013. In: Fire safety in the home. 2011.
- Ploubidis GB, Grundy E. Health measurement in population surveys: combining information from self-reported and observer-measured health indicators. Demography. 2011;48(2):699–724.View ArticlePubMedGoogle Scholar
- Rabe-Hesketh S, Skrondal A. Classical latent variable models for medical research. Stat Methods Med Res. 2008;17(1):5–32.View ArticlePubMedGoogle Scholar
- Skrondal A, Rabe-Hesketh S. Generalized latent variable modeling: multilevel, longitudinal, and structural equation models. Boca Raton: Chapman and Hall/CRC; 2004.View ArticleGoogle Scholar
- Muthen LK, Muthen BO. Mplus user’s guide. 7th edn. Los Angeles, CA. pp. 1998–2012.Google Scholar
- Gorsuch R. Factor analysis. 2nd ed. Hillsdale NJ: Lawrence Erlbaum Associates; 1983.Google Scholar
- Hatfield PM, Staresinic AG, Sorkness CA, Peterson NM, Schirmer J, Katcher ML. Validating self reported home safety practices in a culturally diverse non-inner city population. Inj Prev. 2006;12(1):52–7.PubMed CentralView ArticlePubMedGoogle Scholar
- Yorkston E, Turner C, Schluter P, McClure R. Validity and reliability of responses to a self-report home safety survey designed for use in a community-based child injury prevention programme. Int J Injr Contr Saf Promot. 2005;12(3):193–6.View ArticleGoogle Scholar
- Mason M, Kaufer Christoffel K, Sinacore J. Reliability and validity of the injury prevention project home safety survey. Arch Pediatr Adolesc Med. 2007;161(8):759–65.View ArticlePubMedGoogle Scholar
- Chen L-H, Gielen AC, McDonald EM. Validity of self reported home safety practices. Inj Prev. 2003;9(1):73–5.PubMed CentralView ArticlePubMedGoogle Scholar
- Osborne JM, Shibl R, Cameron CM, Kendrick D, Lyons RA, Spinks AB, Sipe N, McClure RJ. Validity of parent’s self-reported responses to home safety questions. Int J Injr Contr Saf Promot. 2015;1–11.Google Scholar
- Watson MC, Benford P, Coupland CA, Clacy R, Hindmarsh P, Majsak-Newman G, Deave T, Kendrick D. team obotKCSp: Validation of a home safety questionnaire used in a series of case control studies. Inj Prev. 2014;20(5):336–42.PubMed CentralView ArticlePubMedGoogle Scholar
- Watson M, Kendrick D, Coupland D. Validation of a home safety questionnaire used in a randomised controlled trial. Inj Prev. 2003;9:180–3.PubMed CentralView ArticlePubMedGoogle Scholar
- Clamp M, Kendrick D. A randomised controlled trial of general practitioner safety advice for families with children under 5 years. BMJ. 1998;316(7144):1576–9.PubMed CentralView ArticlePubMedGoogle Scholar
- Department for Communities and Local Government: English Indices of Deprivation 2010. https://www.gov.uk/government/statistics/english-indices-of-deprivation-2010. Accessed 27 Feb 2015.