Emergency call dispatchers rely on the information provided by callers to facilitate the delivery of assistance. Authentic callers typically do all they can to cooperate with dispatchers by providing information of relevance to facilitate the fastest possible delivery of assistance. Authentic callers are concerned for the survival and safety of the person in need and, though typically distraught, still attempt to provide information and help the person. There are some callers, however, who utilise emergency calls to verbally ‘stage’ their crime through the call by providing dispatchers with false, inaccurate, or no information in an attempt to conceal or misdirect information about the crime that has occurred. Staged callers’ use of language to purposefully script a misleading narrative is known as a form of verbal staging that can potentially mislead dispatchers, first responders, and investigators about the reality of what has occurred (McKinley and Ferguson 2021). Considering the potential impact of staged emergency calls, it is crucial that police correctly evaluate and interpret emergency calls early on in an investigation. Whilst police are aware of the importance of emergency calls, more research is required to help distinguish the sentiments of guilty and innocent callers. The risk of falsely categorising a call can have major consequences on an investigation, either falsely believing an innocent caller is guilty or believing a guilty caller is innocent. The aim of the current research is not to solve this perennial issue of deception detection directly, but rather to provide a new approach to emergency call analysis that may help police to understand the sentiments of callers towards victims. The emphasis here is on improving emergency call analysis to help develop and facilitate interview strategies that are more effective at information elicitation and establishing rapport based on callers’ linguistic dispositions, rather than simply assigning an individual as guilty or innocent based on their verbal indicators.

If an emergency call is related to a crime or the incident reported is later investigated by police, the emergency call is often referred to as the first interview (Cromer et al. 2018). Whilst emergency calls are not initially taken for this purpose, emergency call recordings can be utilised by police as a means of sourcing an initial voluntary statement. In some instances, the emergency call associated with a police investigation may be one of the purest statements sourced due to dispatchers predominantly asking open-ended questions, such as ‘What is your emergency?’ (Harpster et al. 2009). Owing to the urgent nature of emergency calls, a caller’s language can provide the listener with an insight into the type of relationship or sentiment they have towards the person in need or victim. Therefore, police could use emergency calls to gain a better understanding of a potential suspect’s views or sentiment towards the subject of the call. Whilst many investigators understand the use of emergency calls, providing a more nuanced, scientific analysis may help them use the calls effectively in their investigations, the focus not being on categorising a caller as ‘guilty’ or ‘innocent’, but on understanding their sentiment and how best to elicit information (inculpatory and exculpatory).

More recently, an analytical measure of guilt or innocence has been developed for the sole application to emergency calls, known as the Considering Offender Probability in Statements (COPS) scale (see Harpster 2006; Harpster and Adams 2017; Harpster et al. 2009). The COPS scale, developed by Harpster and Adams (2017), provides a detailed checklist of guilty and innocent indicators that focus on a caller’s language and the content of their call (e.g., ‘Does the caller provide aid to the victim?’). The COPS scale may provide investigators with an initial insight that can assist in progression of investigation, by assisting in the identification of suspects and clearing those who are less likely to be criminally involved. Many law enforcement investigators who are trained in COPS report that they attend caller locations with more awareness and ability to process the scene more effectively. More recently, however, various indicators from the COPS scale have been tested by others (Cromer et al. 2018; Miller et al. 2021; O’Donnell et al. 2022, 2023), who have failed to directly replicate the findings of Harpster et al. (2009). Newer approaches to emergency call analyses have begun to use different linguistic indicators to ascertain caller veracity (O’Donnell et al. 2023). The current research makes a unique contribution by stepping away from the guilty versus innocent debate and instead investigates how else emergency calls may be used more effectively in the preparation of police interviews, the focus being on information elicitation through emotional matching and rapport building in follow-up interviews (Keatley 2023a; Marono 2022), rather than presumptive guilt assumptions.

Police investigations of course do not simply stop investigating a case with the coding of an emergency call as ‘guilty’ or ‘innocent’. Next, the person(s) of interest are formally interviewed. The information gathering interview is where emergency call analyses may be more helpful than a simple ‘guilt’ score. In preparing for and running interviews, many investigators appeal to a suspect’s sense of morality or empathy for a victim, such as, ‘Don’t you think they deserve justice?’ This, however, is an inexpedient strategy if the suspect has a negative attitude, sentiment, or negative linguistic disposition (LD) towards the victim. If the suspect is not concerned about, or indeed dislikes the victim, then appeals grounded in compassion and care will more than likely be ineffective—they may even antagonise the suspect. Investigators could look to a suspect’s emergency call to assist in gathering information since the language used during such a call may reveal the suspects’ thoughts, feelings, attitude or sentiment towards the person in need; this is referred to as their linguistic disposition (LD; Hyatt 2014; Keatley 2023a, b). The LD exhibited towards a person in need exists on a spectrum from positive to negative. Words that reflect care and concern are indicative of a positive LD, whereas disparaging and uncaring words are indicative of a negative LD. Understanding sentiments derived from language has been used in various areas such as advertising (Lin 2022; Tudoran 2019), customer satisfaction (Hu et al. 2017) and public perception (Oscar et al. 2017). No formal scientific study, to the authors’ knowledge, has previously explored the sentiment or LD of emergency callers. Therefore, what is required is a method to identify the LD of emergency callers and to develop an understanding of the differences exhibited between guilty and innocent callers.

Research in detection deception has moved away from attempting to find single ‘tells’ for deception and has increasingly begun to focus on complex, dynamic patterns of indicators in nonverbal communication (Marono et al. 2018, 2017) and forensic linguistics (Keatley et al. 2018; Richards and Keatley 2023; Richards et al. 2023). Whilst many of these researchers have used sequential pattern analysis methods, there are other approaches to temporal patterns (Giebels and Taylor 2009; Taylor 2006). The current research, therefore, will use a different temporal method owing to the need to assess the proximity of a caller’s words and associated actions related more closely or distally to each other. The rationale for analysing the proximity of an emergency caller’s words is based on the premise that certain words and concept are psychologically linked, though spreading activation (Crestani 1997). Spreading activation essentially means that concepts closely related to each other are more likely to be linked with and prime similar thoughts and actions (Crestani 1997). For example, when an offender refers to a victim, they may think of negative connotations and associations; this may then leak out in their words and statements. An analysis of proximity will provide an indication of how closely a caller’s words and actions are linked and the different associations that exist between guilty and innocent callers.

Proximity Coefficients

A measure of proximity coefficients provides an indication of how closely related words or actions are linked, in a complex, temporal analysis (Taylor 2006). Proximity coefficients are based on the premise that behaviours occurring in closer proximity have a temporal link, possibly owing to their common properties or cognitive associations (Taylor 2006). Identifying the proximity of a callers’ words and actions may provide a way of quantifying the relationships of indicators present in emergency calls, allowing for a statistical comparison to take place between guilty and innocent calls. This has a benefit over traditional behaviour sequence analysis (Keatley 2018, 2020) approaches, which only focuses on pairings. Proximity coefficients analyse the entire data string of words from start to end of the episodes and show proximity across all indicators, rather than only lag-one pairings.

A measure of proximity coefficients is achieved by assigning values between 0 and 1 to the indicators present in an emergency call. The initial call is transcribed and coded into a sequence of indicators that occur in succession. A proximity coefficient value of 0 is assigned to behaviours that occur at the furthest ends of the sequence (e.g., A–Z) and a value of 1 assigned to those which occur in immediate proximity (e.g., A–B). For example, in the following sequence, A B C D E, the coefficient between ‘A’ and ‘B’ is 1, whereas the coefficient between ‘A’ and ‘E’ is 0. This is then repeated over the entire group of guilty or innocent callers’ sequences to find which pairings, on average, occur in closer or further proximity. The results of such an analysis provide an indication of which indicators occur more commonly in close proximity. Differences between guilty and innocent callers in terms of their positive and negative LD indicators might provide insight into their underlying sentiments—beyond the simple approach of counting particular words (as is the common approach in sentiment analysis research). A measure of proximity coefficients has previously been applied to a variety of areas, such as crisis negotiations (Giebels and Taylor 2009), police shootings (Porter 2022), and terrorism (Corner and Gill 2019). There is no previous research using proximity coefficients that is related to the analysis of emergency calls.

Present Study

To potentially assist police with emergency call analysis in preparation for information gathering interviews, the current research will explore the LD that callers exhibit towards a person in need. This is the first research, to the authors’ knowledge, to investigate the LD of emergency callers, using proximity coefficients. This research is therefore exploratory and a proof-of-concept approach. As it is not yet known how LD relates to an emergency caller’s guilt or innocence, the current research will explore if the proximity of verbal markers that indicate LD differs between guilty and innocent callers. No formal hypotheses were made due to the novel nature of the research, which is typical of temporal analyses (Keatley 2020; Richards and Keatley 2023). It is, however, likely that there will be a difference in proximity between guilty and innocent callers, taking into account previous research on emergency call analysis (Harpster et al. 2009; Odonnell et al. 2023).

Methods

Sample

The sample consisted of 30 emergency calls, with a known age range of 22 to 61 yearsFootnote 1 (M = 39.68, SD = 11.80): 16 women (6 guilty; 10 innocent) and 14 men (9 guilty; 5 innocent). Emergency calls were sourced from westernised countries, with calls originating from Australia (n = 1), the UK (n = 5), and the USA (n = 24). Whilst slight linguistic differences exist between westernised cultures, their shared communication styles and linguistic features (Altarriba and Basnight-Brown 2022) allow for an accurate comparison between callers. Half (n = 15) of the emergency calls were made by innocentFootnote 2 callers requesting assistance for a victim, and the other half (n = 15) were guilty callers attempting to avoid criminal detection by utilising the call as a means to engage in verbal staging. This sample size and group split is commensurate with other proximity coefficient research (Corner and Gill 2019; Giebels and Taylor 2009; Porter 2022), statement analysis research (Keatley et al. 2018; Richards and Keatley 2023; Richards et al. 2023), and emergency call research (Harpster et al. 2009; Laforest 2012; Lefter et al. 2011).

Data Collection

Full and complete audio recordings of 30 emergency calls were sourced and transcribed from online, open-source platforms, including YouTube, police, and news media sites. There were some instances in which the calls collected had redacted the identifiable details (e.g., name, address, and phone number) of the caller and those related to the call. The redaction of identifiable details was not a cause for exclusion or concern, as the focus of the analysis was on the delivery of this information (e.g., does the caller refuse to provide an address) and not the content (e.g., callers’ address). All calls were related to emergencies and involved a caller providing information about a victim reportedly suffering from an injury, experiencing a medical episode of life-threatening nature, or having died. The decision to include calls from a variety of different circumstances mimics the approach taken by other researchers (Cromer et al. 2018; Harpster et al. 2009), who have grouped calls based on their shared focus of another person(s) health and welfare (as opposed to calls about missing persons and their whereabouts). These emergency calls were real-world, declassified emergency calls; the authenticity of which was supported by official documents (e.g., court transcripts and media releases by police). All emergency calls were sourced from adjudicated cases, in which a caller’s guilt had been proven by law through court outcomes or were otherwise found not guilty through investigation and evidence. If an emergency callers’ guilt or innocence had been subject to public controversy, as evident by news reports, retrials, or ongoing police investigation, their call was excluded from the dataset; this perhaps overly restrictive approach was done to safeguard the data from contamination of misclassified calls. To ensure the quality of the data and that the calls collected were relatively similar for comparison, they were screened against exclusion criteria during collection. The exclusion criteria were as follows: emergency calls that involved more than one caller talking to dispatchers were excluded as it was beyond the scope of the current study to investigate multiple callers. Emergency callers who had redialled emergency services after an initial call were excluded, as it is not yet known if the dynamics of multiple call incidents differ from that of single calls and was beyond the scope of this study. Emergency calls that did not originate from Australia, the UK, or the USA were excluded due to cross-cultural differences and to avoid language issues. Callers within the chosen countries were then screened for English fluency and excluded if not deemed fluent; this was determined by monitoring for broken English.Footnote 3 If exclusion criteria were unable to be verified, the call was excluded as a source of data.

Data Coding

The emergency calls were transcribed and then analysed for indicators of LD and those in the COPS scale. Three coders independently coded calls and agreed on the final code sequences before analyses, which is the standard approach in temporal analyses (Keatley 2020; Keatley and Clarke 2020a, b). A codebook of 125 potential indicators were used to code guilty and innocent emergency calls. The process of coding each call involved reading the call transcript whilst listening to the audio to identify the indicators present and then recording the corresponding codes in the order they appeared. This resulted in a sequence of codes for each call, as shown below (Fig. 1). During this process, it was essential to listen to the audio recording for each call to fully transcribe all aspects of the call (including pauses in speech, stuttering and, false starts); all of which are important to the analysis.

Fig. 1
figure 1

Example of a coded emergency call sequence

Full size image

Indicators of linguistic disposition were developed a priori and a posteriori. A priori codes directly related to a caller’s sentiment were coded for based on literature reviews into sentiment and interpersonal relationships (Lathren et al. 2021; Le et al. 2018), as well as indicators of veracity derived from the COPS scale (Harpster et al. 2009).Footnote 4 After careful reading of emergency call transcripts, any additional indicators noted in the call were added to the codebook to create an exhaustive list for temporal analyses, which is a prerequisite (Keatley 2018, 2020). Indicators of disposition were based on the content of an emergency caller’s conversation (e.g., focus of the call, the type of information provided) and actions (e.g., cooperation with dispatcher, actions performed at scene). Indicators such as ‘address of victim’ were coded and counted on each utterance, leading to the high repetition frequencies observed. This is typical and recommended in temporal analysis research (Keatley 2020; Magnusson et al. 2016; Richards and Keatley 2023). The importance of repeat coding is in understanding the difference between a single indicator (e.g., ‘Help’ or ‘John’) and a burst pattern (Magnusson et al. 2016), which may indicate urgency (e.g., ‘Help, Help, Help’ or ‘John, John, John’). Once calls were coded, backwards translation was conducted (see Keatley 2018, 2020) to ensure that the codes accurately reflected the content of the call. This was achieved in all calls.

Analyses

Proximity coefficients were calculated using software originally developed by Taylor (2006): ProxCalc: proximity coefficient calculator software. This software counts the codes found in sequences and calculates the proximity coefficients in sequences. The mean proximity coefficient values for guilty and innocent callers were reported in separate tables. To interpret the results, a cue (a variable that proceeds another) is located horizontally along one of the rows and then shown in proximity with a sequitur (a variable that occurs sometime after the cue) in a column. The ProxCalc tables show the mean proximity score of the two indicators and provide an insight as to the relationship they shared amongst the guilty or innocent callers’ linguistics. If a code never follows another, it is assigned an undefined value, as represented by ‘—’. If an entire row or column consists of multiple undefined values, it shows that the variable in question commonly occurs at the beginning or end of the calls analysed. If a variable did not exist in a group’s dataset, the corresponding row and column for that table are simply left blank. Note, blank rows and columns are left in the tables to highlight indicators that appear in guilty but not innocent calls or vice versa. For further reading to assist with the interpretation of proximity coefficients, there are a large number of sources (see Corner and Gill 2019; Giebels and Taylor 2009; Keatley 2020; Porter 2022; Taylor 2006).

Results

Reporting on the entirety of the indicators and their associated proximity coefficients would be beyond the scope of comprehension and render end output results uninterpretable.Footnote 5 Therefore, temporal analyses typically reduce complex datasets into manageable, focused outputs (Keatley and Clarke 2021). Whilst the initial analyses were conducted on the full dataset, the output provided here focuses on the theoretically important indicators related to a caller’s LD; the frequency of these indicators can be seen in Table 1.

Table 1 Indicator frequencies
Full size table

Before observing the proximity of indicators, it is important to note the frequency and type of indicators coded between guilty and innocent calls. Of the 15 indicators (shown in Table 1) that were selected to present for proximity, four were not present amongst innocent callers but present for guilty, and two were not present amongst guilty callers but present for innocent. The following indicators were not present in any of the innocent calls, ‘no aid provided to victim’, ‘negative linguistic disposition’, ‘insult/blame victim’, and ‘passive language’, but were all present for guilty callers (n = 8; n = 7; n = 6; n = 5, respectively). The following indicators were not present in any of the guilty calls, ‘positive linguistic disposition’ and ‘complete social introduction’, but were both present for innocent callers (n = 9; n = 1, respectively).

Three of the most frequently occurring codes recorded amongst guilty and innocent calls were ‘address of victim’, ‘coach of victim’, and ‘aid provided to victim’. The most frequently recorded indicator amongst all calls (guilty and innocent) was ‘address of victim’, coded a total of 327 times, appearing in 60% of all calls. ‘Address of victim’ was the most frequent indicator coded in innocent calls (n = 262); however, the most frequent indicator coded amongst guilty calls was ‘aid provided to victim’ (n = 86).

Proximity Coefficients

As is typical in temporal research, once frequencies of indicators have been established, the temporal analyses follow. In the current dataset, proximity coefficient analysis was conducted for all indicators present in guilty and innocent calls. Proximity coefficient diagrams were produced to provide a quick visual overview of the data (see Figs. 2 and 3). The use of heat mapsFootnote 6 applied to Table 2 and Table 3 provides an easier means to identifying indicators of relevance based on their proximity coefficient value. Cells highlighted in darker shades of green signify indicators that occur closer together; those in orange-to-red show indicators that are further apart. To allow a clearer comparison and contrast of guilty and innocent calls, Table 4 outlines the difference between innocent-guilty scores.

Table 2 Average proximity coefficient values of guilty emergency calls
Full size table
Fig. 2
figure 2

Guilty indicator proximity diagram

Full size image
Fig. 3
figure 3

Innocent indicator proximity diagram

Full size image
Table 3 Average proximity coefficient values of innocent emergency calls
Full size table
Table 4 Difference in proximity coefficient averages between guilty and innocent emergency calls (innocent-guilty)
Full size table

When exploring the proximity of indicators that followed a ‘focus on the victim’, guilty callers often followed in close proximity with the indicators of ‘address of victim’ (0.937) or ‘insult/blame victim’ (0.937). Innocent callers, in contrast, were not as proximal to ‘address of victim’ (0.765) and did not share any proximal relationship to ‘insult/blame victim’ as it was an indicator that was not present in the innocent call dataset. Indicators that were proximal to ‘focus on victim’ amongst innocent callers include ‘aid provided to victim’ (0.810), ‘bleeding comments’ (0.863), and ‘coaching of victim’ (0.860).

An utterance indicating ‘aid provided to victim’ was followed by various indicators between guilty and innocent calls. Guilty callers showed an association between ‘aid provided to victim’ and ‘negative linguistic disposition’ (0.806) and a ‘defendant mentality’ (0.229), whilst innocent callers showed a proximity between ‘aid provided to victim’ and ‘positive linguistic disposition’ (0.595) but no proximity to ‘defendant mentality’. Of the indicators that were present amongst innocent callers, ‘bleeding comments’ (0.880) and ‘coach of victim’ (0.896) were of the closest proximity following ‘aid provided to victim’.

The proximity coefficients of indicators that followed an ‘acceptance of (victims) death’ for guilty callers were followed closely with the use of ‘distancing language’ (0.846). In contrast, when innocent callers expressed an ‘acceptance of death’, an indefinite value is reported (as represented by ‘—’), which indicates no proximal relationship exists between the two indicators. Of the indicators that did follow an ‘acceptance of death’ amongst innocent callers, the most proximal indicator was related to remarks that revealed a ‘positive linguistic disposition’ (0.847) towards the victim.

Discussion

The primary aim of the current research was to provide a new approach to emergency call analyses, using an established temporal analysis approach. The purpose of this was to show that emergency call analyses can provide a useful source of information for follow-up investigations that are scientifically, psychologically informed. The linguistic disposition (LD) of callers was coded alongside multiple existing emergency call indicators. Proximity coefficient analysis was then used to map the temporal links between indicators in guilty and innocent calls. Results provide initial support for the role of LD analysis in emergency calls, using temporal analyses.

Prior to analysing proximity, initial frequency counts showed a difference in the types of indicators present amongst guilty and innocent calls. Whilst the results should not be exaggerated to suggest that guilty and innocent callers are fundamentally separable based on indicators, there were instances in which indicators only appeared in guilty or innocent calls. Four indicators were present in guilty calls but absent in innocent, whilst another two indicators were present in innocent calls but were absent in guilty. The presence of different indicators between guilty and innocent callers is consistent with the wider literature on deception detection, as there are certain words and grammatical structures expected amongst those who are deceptive and truthful (Schafer 2007; Harpster et al. 2009). Of particular importance in the current research approach, negative and positive LD was only seen in guilty and innocent callers, respectively. It would be unwise to overstate the interpretation and meaning of these results here. The take-home message here is not that caller’s guilt or innocence can be determined by simply looking at these indicators. Indeed, the fact remains that a vast majority of the same indicators occurred in both guilty and innocent calls, albeit to different frequencies. However, these results do provide some evidence for differences in disposition between guilty and innocent callers.

Whilst the frequency count results show that guilty and innocent callers differ in the type and frequency of indicators, alone, it is not necessarily enough to accurately understand the LD a caller exhibits in an emergency call. In this way, the present study is not a critique or criticism of past research, as a novel focus was taken. Past research that has attempted to discriminate between guilty and innocent calls with the use of single indicators has been unable to find a clear difference between caller types (Cromer et al. 2018; Miller et al. 2021; O’Donnell et al. 2022). The inability to identify associations of single indicators between guilty and innocent callers provides support for the development of a measure that goes beyond identifying the presence or absence of individual indicators. Whilst the identification of individual indicators is important and may provide some initial insight, it may be that the sum is greater than the value of its parts (Clarke and Crossland 1985; Keatley 2018) and that the appropriate approach is to look for clusters or patterns of indicators. This approach has previously been used to explore the verbal and non-verbal cues to deception in public appeals in missing or murdered person cases (Whelan et al. 2014).

Rather than attempting to ascertain the guilt or innocence of a caller, which may cause investigative issues if false assumptions are made, the current research provides a novel pathway into understanding a caller’s sentiment towards the subject of the call. By identifying a caller’s sentiment at the emergency call stage, police investigators are provided with a source of information that could assist them in preparing interview strategies. Following an emergency call, investigators may interview the caller. To assist investigators in developing an interview strategy that is more nuanced and perhaps better suited for the caller, they could look to the LD they exhibited in their call. For example, callers who express a negative LD towards the person in need are unlikely to feel care or concern for them in an interview. This suggests that appeals targeting the callers concern for the victim, such as ‘it is clear that you cared quite deeply for them’, are unlikely to elicit information from them as it does not resonate with their LD. Instead, investigators could make an appeal that aligns with or more closely matches the suspect’s negative LD. These suggestions as to the application of LD clearly require further research; it does, however, highlight how an analysis of emergency caller LD could be applied in the development of interview strategies.

Rapport building is a central and fundamental part of interactions, especially in information elicitation and sharing (Marono 2022). Therefore, the current research highlighting the importance of LD may also assist investigators with building rapport in investigations. To assist police investigators in preparing interview strategies that will aid in building rapport, they could look to ‘matching’ the LD an interviewee exhibited during their emergency call. Previous research has shown that the matching or imitation of another’s behaviour (Marono 2022), language (Richardson et al. 2014) or emotions (Fischer and Hess 2017), otherwise known as mimicry, can increase feelings of closeness and encourage cooperation. Mimicry has previously been explored within the context of police investigations to assist in building rapport (Abbe and Brandon 2014; Brimbal et al. 2019; Richardson et al. 2014), but researchers are yet to explore the matching of sentiment, as derived from the LD exhibited in emergency calls. This approach to emergency call analyses, interpretation, and follow-up investigation may provide police with a more cautious approach to information elicitation, rather than presuming guilt too early.

Whilst there are several strengths to the current research, there are also several limitations that should be noted. First, the sample is relatively small, though commensurate with other statement analysis (Keatley et al. 2018; Richards and Keatley 2023; Richards et al. 2023) and emergency analysis (Harpster et al. 2009; Laforest 2012; Lefter et al. 2011) research in the field. Now, initial support has been found for the approach and analyses; further refinement of the indicators on larger samples is required. For example, ‘aid provided to victim’ in the current dataset was coded when actual aid was clearly provided. There is of course nuance to those times when aid may only be implied (like asking about aid, rather than providing aid). These nuances could be further explored.

This research should be understood as a proof-of-concept study, both in terms of its basis and its application to real-world cases. What is clearly needed is further research with a larger sample size to corroborate the findings and to explore the relationships between indicators and interview strategies. A strength of the current research is that the analyses allow for complex dynamics to be explored and understood in a way that can provide investigators with new approaches to developing interview strategies that increase rapport and information gathering. This non-adversarial approach is a positive direction for research that has typically adopted a more ‘deception detection’-oriented approach (Cromer et al. 2018; Harpster et al. 2009; Miller et al. 2021; O’Donnell et al. 2022, 2023). Whilst the current research offers a potential new approach with some preliminary support, further research is required before direct application to real-world ongoing cases.

Conclusions

The current study took a novel approach by not focusing on the guilt versus innocence coding of emergency call analysis and instead focused on analysing the underlying psychological sentiment a caller exhibits. Considering the issues inherent in previous research to identify single indicators, a holistic, dynamic system approach was taken to measure the complex proximity patterns across indicators present in emergency calls. Initial support was provided for analysing the proximity of indicators exhibited in guilty and innocent emergency calls. This research provides the foundational first step towards understanding how emergency call analysis with a focus on LD can provide insights that may be used to inform police investigations. Caution is recommended before applying these results in real-world cases before further research and support are found.