A Study of Differences in the Perceived Risk of Attaining a Residential Fire Injury

Certain groups (e.g., those of higher age, lower educational attainment, lower income, living alone, immigrants, etc.) have a greater risk of residential fire mortality. Previous research has also shown that individuals belonging to high-risk groups have generally lower levels of fire protection, and it has been suggested that this is due to a lower risk perception in this group. As such, this study investigates how the perceived risk of being injured in a residential fire varies in the Swedish population. The results show that risk perception varies in the Swedish population depending upon sociodemographic factors. When the different sociodemographic factors are controlled against each other, women, individuals with a low educational level, individuals living in rural communities and individuals born outside of the Nordic countries consistently experience their risk to be higher. With the exception of women, the results show that high-risk individuals have a high risk perception. These results are important as they indicate that it is not a lack of risk awareness that is the reason why high-risk groups are less inclined to implement fire safety practices.

*Correspondence should be addressed to: Finn Nilson, E-mail: finn.nilson@kau.se Fire Technology, 59,[1789][1790][1791][1792][1793][1794][1795][1796][1797][1798][1799][1800][1801][1802][1803][1804]2023 Ó 2023 The Author(s) Manufactured in The United States https://doi.org/10.1007/s10694-023-01410-x However, it is also well-known that fatal fires can be hindered at several stages [14], and several studies have shown that interventions such as smoke alarms, education or multi-facetted programs are effective [3,[15][16][17], if they are adjusted to different sociodemographic groups [16,18]. On an aggregate level, however, highrisk groups seem to be less inclined to employ protective measures [19]-according to some [20]-due to their low perceived risk of their situation. In other words, given their low or faulty perception of their individual risk, they refrain from applying preventative measures. If high-risk groups have a faulty perception of their risk, this could explain why sociodemographic groups with high risk of fire mortality to a lesser degree have suitable risk preventative measures implemented in their homes [19] and are to a greater extent reliant on societal protection in the prevention of fire fatalities [21][22][23][24].
Risk perception and fire safety has been approached in several different ways in the fire safety literature. The Protective Action Decision Model (PADM) [25] is a commonly invoked theory in risk perception of fire safety, particularly in helping to understand human behaviour and risk perception in relation to evacuation [26,27]. An inherent assumption of PADM is that individuals often have erroneous risk perception and that society has a responsibility to correct such misconceptions [25]. However, although several studies have indicated that individuals may not entirely assess risk correctly, [28][29][30], especially regarding risks that are rare, if they are promoted in the media or evoke dread [29], this does not mean that individuals are unaware of their risk or unable to determine their own risk. On the contrary, data suggests that individuals tend to have quite accurate perceptions of their mortality risk-at least in relation to relatively common hazards [31][32][33]-with the exception of men consistently assessing their risk as lower compared to women despite having a higher objective risk [34].
Why certain risks can be understood correctly and other faultily can in turn be understood through the Protection Motivation Theory (PMT) [35]. The basis of PMT is that fear is constructed through three elements: (1) the noxiousness or severity of an event; (2) the probability of an event occurring if no protective measures are taken or behaviour is adapted; and (3) the efficacy of a response to reduce or eliminate the noxious event [36]. Therefore, as knowledge, control and self-efficacy are crucial ingredients in risk perception, common hazards that originate close to the individual and that are easily dealt with are more often understood correctly, compared to fuzzy and uncontrollable hazards. Residential fires could be viewed as both rare and uncontrollable events as well as common occurrences with smaller residential fires being relatively common [37] although most people will not experience a serious event-not least given the sociodemographic differentiation in risk [38].
As such, it is not clear-neither from the current literature nor the theoretical understanding of risk perception-whether there are sociodemographic differences in fire risk perception. This study will therefore investigate how the perceived risk of being injured in a residential fire varies in the Swedish population and how this perceived risk varies depending upon different sociodemographic factors. Also, the paper will investigate which factors are the most important and if these factors correspond to the known sociodemographic risk factors underlying fire mortality.

Materials
This study uses a dataset from the so-called Tillitsbarometern [39], a population survey that studies variations in trust among a random sample of individuals aged 18-85 living in one of 45 municipalities (deliberately chosen to represent a maximum variation in factors that are assumed to affect different forms of trust) around Sweden. Data has previously been collected in 2009 and 2017. In the autumn of 2020 data was collected a third time and several risk-related variables were added to the questionnaire, including the question of how the individual perceived their own risk of being injured in a residential fire. The wording of the question, specifically asking about the risk of injury and not the risk of residential fire, is due to two perspectives. Firstly, given that the overall survey is related to the trust in others (both institutions and other people), simply asking about the perceived fire risk would not necessarily include dimensions of potentially needing help from others. Secondly, although all residential fires are unwanted, a large majority of residential fires do not lead to injury and do not require assistance from neighbours nor rescue services [40]. The overarching issue at hand is to minimise injuries and fatalities due to fires and help ascertain why the fire fatality trends are no longer decreasing [41]. As such, if one can handle a fire by oneself, the risk of fire may be considered great, but the risk of fire-related injury is considered small. Consequentially, although the question does convolute two different aspects of fire prevention, i.e., both fire and the effect of the fire [42]-which is somewhat problematic-the question focuses on injury risk as it is injury that is predominantly undesirable.
Alongside questions on trust, risk and attitudes, the survey includes several sociodemographic variables as well as being linked (through the highly reliable Swedish Personal Identity Number [43]) to official registries regarding age, sex, income, etc. The total number of responses of the survey was 13,667.

Statistical Methods
The statistical analysis in this paper was performed in several steps. Starting with how the perceived risk of being injured in a residential fire varies in relation to sociodemographic characteristics, a Kruskal Wallis H test was performed. This type of analysis was chosen to provide a general understanding of how different groups vary in their risk perception. A non-parametric test, rather than an ANOVA, was performed due to the ordinal character of the dependent variable, as well as lack of normality in the distribution of this variable (see Online Appendix 1 for distribution of all included variables). The significance level was set to 0.05. To determine whether characteristics differed significantly from each other within each variable, a post-hoc test (Dunn-Bonferroni test) for multiple comparisons was performed, as is suitable for a Kruskal Wallis test [44]. The significance level of the post-hoc test was thus adjusted for a Bonferroni correction for multiple tests.
Second, to investigate which characteristics had the greatest impact, compared to the others, in affecting the perceived risk of being injured in a residential fire, a multivariate regression analysis was performed, where one variable was added in each new model. Due to the non-normal distribution of our dependent variable, we chose to perform a logistic regression, after converting our dependent variable to a binary variable that indicates whether respondents perceived a risk of getting injured in a residential fire or not. The alternatives ''Very small'' and ''Fairly small'' were combined into one category (0), and the alternatives ''Neither/nor'', ''Fairly large'' and ''Very large'' were combined into another category (1). The decision to include the neutral option in the latter category was based on an assumption that most people would assess their risk as low, and by choosing the neutral option the respondent would be open to there being some risk. In order to illustrate the unique effect of each variable, we chose to perform a stepwise logistic regression analysis. The significance level of the regression analysis was set to 0.05, as is the norm. However, in accordance with the current discourse surrounding p values [45], we have also added Akaike Information Criterion (AIC) for each new variable added in the regression to further describe the explanatory power and relationships between each specific model. Third, to further validate our findings in the stepwise regression model, we also conducted an automated forward (LR) stepwise regression analysis.

Results
As seen in Table 1, several statistically significant differences in perceived risk were seen in terms of sociodemographic groups. Although there was no statistically significant difference between people living alone (single households) or households with more than one person (p = 0.103), in all other topics significant differences are seen. Men perceived their risk to be significantly lower than women, as did younger age groups compared to older adults. Studying the effect size (epsilon2), gender and education stand out, as do age, income and occupation with significantly lower perceived risk amongst men, higher educated, younger, and higher income individuals. Students and employed/business owners perceive their risk as lower compared to those off work.
In terms of multiple comparisons, the Dunn-Bonferroni test found significant differences in perceived risk between respondents born in the Nordic countries and the rest of Europe (adj. p < 0.001) with those born in the Nordic countries perceiving their risk as lower. There was no significant difference between respondents born in Europe (including Nordic countries) and born outside of Europe. Furthermore, the post hoc tests found that perceived risk was significantly different between high income and medium income, H = 800.344, adj. p < 0.001, and between high income and low income, H = 995.024, adj. p < 0.001, with a higher income corresponding to a lower perceived risk. There was no significant difference between low-income respondents and medium-income respondents.
Comparing different occupations, the Dunn's post-hoc test found that perceived risk was significantly higher amongst those retired compared to the category Other, H = 958.845, adj. p < 0.001. Similarly, students (H = -1352, adj. p < 0.001,), home workers, those on sick leave or unemployed and retired (H = -336.186, adj. p 0.006,), and employed or business owners (H = -889.850, adj. p < 0.001,) all perceived their risk as significantly lower compared to retired individuals. Finally, students experienced their risk as significantly lower In terms of educational level, the post-hoc test revealed significant differences between all levels of education, with perceived risk decreasing with each level of further education. The largest difference was between high and low educational level, H = 1347.068, compared to high with medium level of education, H = 532.673, and when comparing low and medium education levels, H = 814.395.
Comparing residential types, the perceived risk was significantly lower for people living in an owned apartment compared to those living in a rental house or apartment, H = -481.312, adj. p < 0.001. Likewise, those living in an owned house, townhouse, or farm perceived their risk as lower compared to those living in an owned apartment, H = 506.325, adj. p < 0.001.
Finally, in terms of socio-geographic categories, significant differences were found between all groups, with an increase in risk perception in each step towards rural areas. The largest difference was found between those living in urban and those living in rural areas, H = 922.629, adj. p < 0.001, where those living in urban areas perceived their risk as lower. Whilst less of a difference, those living in urban areas also perceived their risk as lower compared to those living in ''close to urban'' areas, H = 364.491, adj. p 0.015. Similarly, those living in ''close to urban'' areas perceived their risk as lower compared to those living in rural areas, H = 558.138, adj. p < 0.001.
Given the large number of statistically significant results, a stepwise logistic regression was performed, with a total number of nine models. As is clear in Table 2, in which odds ratio with confidence intervals (** = p set to 0.01, * = p set to 0.05) are presented, many of the differences (or lack thereof) seen in Table 1 remain. For instance, the difference between men and women (with women perceiving their risk as higher) remains strong in all nine models, regardless of how many other sociodemographic factors are controlled for, with an odds ratio of 1.9 in the full model, i.e., the likelihood of women reporting higher perceived risks were almost twice as high as for men. Education also remains an important factor. Socio-geography remains significant, where people living in urban areas have a lower risk perception compared to those living in rural areas, even when controlling for other sociodemographic factors. Whether you live alone or not does not appear to be relevant in either our analyses. Being born in another European country, outside of the Nordic region, seems to increase the odds for perceiving a risk of fire injury. However, in contrast to the Kruskal Wallis analysis, we see that being born outside of Europe is less of a relevant factor when controlling for other variables. Finally, in model 4, we see that the odds ratio for the older age group decreases when we add occupation to the model.
For each model, the fit relative to other models is represented by an AIC value. Comparing the AIC values of the different models, it is clear that adding variables will always increase the AIC and the model with the lowest AIC is the model that only includes one variable. This could be interpreted as there are not one or a few variables included in this model that independently strongly out-weigh others. As such, it can be assumed that although there are variables that show significant dif-   ferences in perceived risk, each variable is not more important than others in terms of the explanatory power. This is further strengthened when looking at the changes in Nagelkerke values between each model, i.e., no individual variable or combination of variable is decidedly more important than the others. This tendency can be confirmed further in the automated stepwise analysis (Table 3), where only one variable was eliminated.

Discussion
As is clearly seen in this paper, individuals' perceived risk for sustaining a fire-related injury varies in the Swedish population and is affected by sociodemographic characteristics. Starting with perceived risk and its variation across different subgroups, several interesting findings exist. First, this study shows that men seem to perceive their risk of being injured in a residential fire to be lower than women, despite being overrepresented in fire fatalities [2,3]. However, apart from this finding, practically all results show that risk perception is relatively well-aligned with known risk factors. Individuals with known risk factors, such as higher age, lower income and education, being born outside of the Nordic countries, and living in rural areas, all perceive their risk to be higher-a perception that largely mirrors their actual risk in a Swedish setting [2,13]. An important finding in this paper is also that although strong significant differences are seen in relation to several variables, the explanatory power in the different models is low. Whilst this does not entail that the identified significant differences are not relevant, it does raise the question of whether there are other (non-identified) factors that explain differences to a greater degree, or whether the different variables are so intertwined that one cannot separate them from each other? We would argue that the second explanation is more likely, not least when taking the results of the automated forward stepwise regression model into account. Hypothetically, individuals who perceive their risk as high have so many of the risk factors that they cannot be separated from each other. Similar findings regarding multi-risk individuals has been shown in studies on fire mortality when Excluded variable: single household many interlinked different sociodemographic risk factors (rather than a single factor or only a few) increase the risk of perishing in a fire [13]. Regardless, with such similarities between risk perception and actual risk, there are important and interesting ramifications from a prevention perspective. Previous studies have found a significantly lower use of preventative measures or practices amongst ethnic minority families [46][47][48][49], single-households and low income families [50], individuals with a lower educational level [51,52] as well as those living in socially deprived areas [53,54], i.e. background factors that are similar to the groups found in this study to have a higher risk perception. The findings in this study therefore clearly question the hypothesis that fire protection is to a greater degree absent in certain groups due to faulty risk perception and that high risk perception leads to improved safety [20]. In fact, fire morbidity risk seems to largely be understood correctly, thereby indicating that high-risk individuals, who perceive their risk as high, are (on an aggregate group level) knowingly refraining from fire prevention practices for other reasons.
These findings therefore indicate that many individuals may be aware of the fact that they require help during a fire. As such it is even more surprising that safety practices are often low in the same groups [19], as the results indicate that individuals' assessment of their situation may be considerably better than could be assumed. Consequentially, this leads to two conclusions. First, that other factors may hinder individuals from implementing safety precautions despite knowledge of their situation, and second, that it does not seem to be a lack of knowledge or insight that hinders prevention.
Starting with the second point, information regarding fire safety has often been a proposed method in the hope of reducing fire fatalities. Although the effectiveness of information in injury prevention has been questioned severely [55], the results in this study on aggregate risk perception further illustrates the lack of theoretical effectiveness of trying to convince individuals to change behaviour by increasing their fear of the situation [56]. In fact, this study indicates that individuals (on an aggregate level) seem relatively competent in determining their risk profile meaning that societal fear-generating is most likely futile if the purpose from authorities is to increase fire protection measures by increasing fear.
Returning to the first point, several factors could be the cause of a lack of action despite knowledge. For example, financial reasons have been known to hinder older adults in improving their fire safety [57]. However, in clinical trials the distribution of free smoke alarms did not reduce fire-related injuries, predominantly because the smoke alarms were not maintained or installed [58]. In turn, this raises the issue of risk prioritization, i.e., that although individuals are aware of the risk of fire-related injuries, other issues are more pressing in everyday life. Such results have previously been observed in relation to fall-related injuries [59] and may be relevant also for fire safety. Regardless, the results clearly point towards the importance and prioritisation of passive interventions [60] in order to reduce fire-related injuries.
Although the results in this study are important and, we believe, sound, there are some limitations that need to be addressed. First, there are some issues with the data collection. Questionnaires almost always have a certain selection bias in terms of which groups answer. Whilst attempts have been made to minimise this through detailed analysis of non-responses, one must view the results thereafter. Similarly, the data collection is not national. Instead, the survey used a purposeful selection of municipalities that were deemed representative in terms of the Swedish population's differences in trust. As such, we believe that the results most likely represent the Swedish population although this cannot be entirely certain. Furthermore, some limitations concerning the dependant variable should be mentioned. First, the formulation of the risk perception question is not based on a scientific definition, which could make it harder to compare the findings of the study to previous work. Second, as mentioned previously, the question entails two aspects of fire prevention; both fire and the effect of the fire, which could be problematic. Third, using a single item will always bring an amount of reliability uncertainty, as we cannot know that the respondents understand the question in the same way. Fourth, the use of nonparametric tests limits the interpretation of the results, as no linear relationships can be claimed. Dummy coding the dependant variable for the logistic regression was performed for an easier interpretation of the results, but we understand that the choice to include the neutral option in one of the two categories could be discussed as arbitrary. However, despite the challenges of nonparametric analyses and other limitations mentioned above, we believe that our study shows interesting results worthy of further discussions within the field of fire safety.

Conclusions
This study shows that risk perception varies in the Swedish population depending upon sociodemographic factors. Practically, the results indicate that high-risk groups are relatively aware of their risk profile. These results are important as they indicate that it is not lack of knowledge or risk awareness that is the problem in terms of why high-risk groups are less inclined to implement fire safety practices.

Declarations
Competing Interests The authors have no competing interests.

Open Access
This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat ivecommons.org/licenses/by/4.0/.