1 Introduction

The continuous enhancement of road safety is considered as one of the primary goals of both the World Health Organization (WHO) and the European Union (EU) [7, 24]. Consequently, it is also a priority for Hungary to increase the level of traffic safety [10].

Moreover, advanced driver assistance systems and automated vehicle modules have excellent safety enhancing potentials, in line with international development goals [5]. Solely by allocating more resources to safety in the development process of modern vehicle and traffic systems, accident risk is continuously decreasing [1, 9, 15].

Despite diminishing accident risks, the deep analysis of occurring accidents will still remain of utmost importance, so that their exact causes could be revealed, and thus the necessary interventions and development directions could be determined. Furthermore, a detailed analysis of accident causes is also indispensable for answering the questions of liability in traffic accidents [6]. As a consequence, the collection of data on the causes of road collisions and their thorough analysis aiming to draw conclusions and to determine appropriate measures are also listed as important goals in the Valletta Declaration on Road Safety [5]: Points 3, 8 g).

The degree to which the causal relations of accidents and consequently, the liability of parties can be assessed differs for various accident types [12, 22]. The complexity of the causal chain increases as the automation level of vehicles is enhanced. Consequently, the role of the human factor in the driving process is to diminish, while the role of other factors is expected to increase. This may lead to a regrouping of liability relations (e.g. liability of the manufacturer, the supplier) [18].

The problem of liability is closely connected to accident reconstruction: based on well-established accident reconstruction calculations, liability can be determined with high certainty. The higher the quantity and quality of basic accident data are, the more assessable accident processes will be [4].

The present study explores the effects of basic data available for calculations and of data recording technologies on the accident reconstruction process, and thus on the possibility of determining liability. It is examined how assessability is expected to improve due to the introduction, spread, and wide application of certain data recording technologies. The types of data recording technologies required to ensure the expected level of assessability are also determined.

After reviewing the relevant literature, the methods followed in this study are described, and the accident database [21] is introduced. The accident types are illustrated by example accidents. Then the influence of data recording technologies on the assessability of accidents in our database is analyzed.

We have developed a methodological framework to evaluate the effects of future changes in accident forms and data recording technologies on accident assessability. Accordingly, the new method makes it possible to investigate, e.g., new forms of accidents arising with the appearance of autonomous vehicles and the effects of new data recording technologies on accident assessability [17].

2 Data and methods

This section describes the structure of the database used for our analysis and the accident types. The statistical methods used are introduced. The aim of the analysis is to reveal how various data recording technologies might influence the assessability of road accident causes and processes. The following flowchart depicts the methodological steps of the present research (Fig. 1).

Fig. 1
figure 1

Steps of the methodology presented here

Full size image

2.1 Accident database

The databased used here was compiled from the data of 124 accidents which we had examined earlier as forensic experts [21]. The data in the database resulted from an in-depth analysis of the given accidents (full accident reconstruction, expert evaluation, simulation calculations, etc.), which in this form can be considered unique in this research field. Of course, more data would lead to a more accurate conclusion. At the same time, the applied statistical tests can also be used to comparisons of small sample sizes, so the conclusions drawn can be considered reliable.

2.1.1 Accident assessability

Collisions were grouped according to how much the accident process can be reconstructed and how well the cause of the accident and the liability for it can be determined.

An accident process can either be

  • fully assessable (can be reconstructed to the required degree);

  • or not or just partially assessable (cannot be reconstructed to the required degree).

The cause of the accident and liability could be

  • determined;

  • or not determined.

It must be highlighted that it is not the competence of the technical forensic expert to determine liability. From a technical point of view, liability is assessable if the cause of the accident can be determined, and a mistake in driving technique of the parties could be identified as a cause. Based on the above two parameters, accidents were put into one of the following four categories (Table 1).

  • The assessability of a given accident is Level 1, if the accident process could not be reconstructed, or could be reconstructed only partially, and the cause of the accident or liability could not be determined.

  • The assessability of a given accident is Level 2, if the accident process could be reconstructed to the required degree, but the cause of the accident or liability could not be determined.

  • The assessability of a given accident is Level 3, if the accident process could not be reconstructed, or could be reconstructed only partially, but the cause of the accident or liability could be determined.

  • The assessability of a given accident is Level 4, if the accident process could be reconstructed, and also the cause of the accident or liability could be determined.

Table 1 Accident assessability levels
Full size table

2.1.2 Analyzed accident data recording technologies

This research investigates the assessability of accidents not only by looking at data collected in the traditional ways, but also by testing how the assessability level would change if data collected by EDR technology were available. The availability of data recorded by highly advanced data recording technologies applied in highly automated vehicles was also tested. In addition to data types of the DSSAD (Data Storage System for Automated Driving—video recordings, location, speed, and/or surroundings of a vehicle; [8, 20], further data types were considered which are required for controlling highly automated vehicles, e.g. the data flow between the electronic control units (ECUs) have also been considered. This third data category, which is related to automatic vehicles (AVs), is labelled EDR+ in this article. The three categories are discussed in detail below. Additionally, we also checked what kind of data exceeding the EDR+ category would have been required to reach Level 4 assessability for the given accident.

In sum, the data available for the accidents analyzed were recorded using the traditional data recording technologies. After analyzing the assessability of each accident, it was determined how assessability values would change if more advance data recording technologies (EDR, EDR+, respectively) had been used. Concerning the traditional (data recording technologies) category, in addition to data recorded at the accident scene by the police, data obtained later in the police investigation process (e.g. witness testimonies) were also considered.

2.1.3 Accident types

The 124 accidents analyzed here proved that the assessability of road collisions require categories that make it possible to separate factors that fundamentally influence assessability, such as: accident cause (e.g. roadway departure), location (e.g. crossroads with traffic lights) or the movement of the vehicle (turning or going straight). Thus the categories were established from the perspective of assessability. Naturally, not all theoretically possible accidents can be fit easily into this categorization, but the overwhelming majority of cases can be assigned to one of the following types [13, 19].

  1. 1.

    Destabilization or roadway departure of a vehicle causes the accident.

  2. 2.

    A vehicle turning left collides with a vehicle overtaking it.

  3. 3.

    A turning vehicle collides with another vehicle going straight.

  4. 4.

    Head-on collision occurs when a vehicle goes over to the lane serving the opposite direction.

  5. 5.

    At least one of the affected vehicles changes the lane immediately prior to the accident or during the accident process.

  6. 6.

    A rear-end collision occurs when a vehicle crashes into the one in front of it, or at least one of the vehicles is backing.

  7. 7.

    For collisions in crossings with traffic lights, the colour of the lights at the time of the accident must be determined.

The aim of the categorization was to determine the major contributory factors that influence the assessability of a given accident. Although generally a given accident could be assigned to more than one category, each collision was put into the most characteristic group. Thus disjoint sets were created, each accident was assigned to one single category.

Explanatory figures and descriptions of the investigated types of accidents can be found in “Appendix 1”.

2.2 Methodology

The examined data recording technologies (traditional, EDR, EDR+) are denoted T0, T1, and T2, respectively. As a starting point, all accidents in the database were assessed based on data recorded by traditional techniques (T0). For the next steps, it was supposed that EDR (T1) and also EDR+ (T2) data were available. The accident process was determined again with the extended data sets, and the change in the level of assessability was determined.

The assessability of accidents was compared in several ways. For the first three tests, a pairwise comparison of independent, non-normally distributed samples was carried out, by applying the Mann–Whitney Test [14, 16].

$${U}_{1}= {n}_{1}{n}_{2}+\frac{{n}_{1}\left({n}_{1}+1\right)}{2}-{R}_{1}$$
(1)
$${U}_{2}= {n}_{1}{n}_{2}+\frac{{n}_{2}\left({n}_{2}+1\right)}{2}-{R}_{2}$$
(2)
$$Z= \frac{U-\frac{{n}_{1}{n}_{2}}{2}}{\sqrt{\frac{{n}_{1}{n}_{2}({n}_{1}+{n}_{2}+1)}{12}}}$$
(3)

where \({n}_{1}\), number of elements in Sample 1; \({n}_{2}\), number of elements in Sample 2; \({R}_{1}\), sum of the ranks in Sample 1; \({R}_{2}\), sum of the ranks in Sample 2; \({U}_{1}\), test function value for Sample 1; \({U}_{2}\), test function value for Sample 1; \(Z\), value of function approaching the normal distribution.

In Figs. 2, 3, 4, 5, 6, 7, 8, 9 below illustrate what categories were compared in the tests. The tests make it possible to reveal how assessability changes for a given accident type compared to other categories if data recording technologies develop. In the figures, individual columns represent the assessability of a given accident category, the colours encode the data recording technology, which is also marked by labels T0, T1 and T2 (denoting traditional, EDR and EDR+ technologies, respectively).

Fig. 2
figure 2

Comparison of the assessability of one accident type to that of all the others, data are recorded traditionally. (Red—examined accident type, data recorded by T0; yellow—all other accident types, data recorded by T0.)

Full size image
Fig. 3
figure 3

Comparison of the assessability a given accident type, with EDR data recording and the average assessability of a type in the case of traditional versus EDR technologies. (Dark green—examined accident type, data recorded by T1; yellow—other accident types, data recorded by T0; light green—other types, data recorded by T1.)

Full size image
Fig. 4
figure 4

Comparison of the assessability a given accident type, with EDR+ data recording and the average assessability of a type in the case of traditional versus EDR+ technologies. (Dark blue—examined accident type, data recorded by T2; yellow—other types, data recorded by T0; light blue—other types, data recorded by T2.)

Full size image
Fig. 5
figure 5

Comparison of the assessability of data recording technologies for a given accident type. (Red—T0; dark green—T1; dark blue—T2.)

Full size image
Fig. 6
figure 6

The assessability of accident types according to data recording technology with Kruskal–Wallis Test. (Yellow—T0, green—T1, blue—T2)

Full size image
Fig. 7
figure 7

Comparison of pairs with a significant difference with Mann–Whitney Test. (Yellow—T0, green—T1, blue—T2)

Full size image
Fig. 8
figure 8

The assessability change (Δ) of accident types according to the development in data recording technology with Kruskal–Wallis Test. (Yellow—T0, green—T1, blue—T2)

Full size image
Fig. 9
figure 9

The comparison of significantly different changes (Δ) with the Mann–Whitney Test. (Yellow—T0, green—T1, blue—T2)

Full size image

For the comparisons, a given subset is chosen from the databased (e.g. Accident Type 1, T0 data recording technology), and assessability is compared to the relevant total set (e.g. all accident types, T0 data recording technology) and also to the relevant complementary set (e.g. Accident Types 2–7, T0 data recording technology). In Figs. 2, 3, 4, 5, 6, 7, 8, 9, the dashed line represents comparison with the relevant total set, while the ellipsis illustrates comparison with the complementary set. In the course of our investigations, following the general methodology, we compared the investigated accident category with the group representing the population. During the comparison, the group representing the population may include all accident categories, including the investigated accident category (total set), or all accident categories without the investigated accident category (complementary set). To avoid drawing false conclusions, we performed both types of comparisons in our analyses.

At first, the assessability of the different accident types was compared to that of all the other accident types, given that data are recorded by T0.

Then, we compared the assessability of the different accident types concerning the average assessability for T0 and T1 techniques, respectively.

In the next test, we compared the assessability of the different accident groups concerning the average assessability of T0 and T2 techniques, respectively.

In the next step, we analyzed the deviation between T0 and T1, T2 data recording technologies by accident types.

The pairwise comparison of connected groups with non-normal distribution was carried out with the Wilcoxon Test [23]: value pairs of the samples are matched with the signs of the difference between the sample value pairs, and sample values are summed, distinguishing between negative and positive values (\({W}^{+}{, W}^{-}).\)

$${W}^{+}=\sum_{{Y}_{i}{-X}_{i}>0}{R}_{i}$$
(5)
$${W}^{-}=\sum_{{Y}_{i}{-X}_{i}<0}{R}_{i}$$
(6)
$${W=W}^{+}{-W}^{-}$$
(7)

In order to rank the assessability of accident types, a group comparison was carried out for the various data recording technologies (T0, T1, T2). In order to diminish the probability of a Type I Error arising from the analysis of multiple hypothesis, the results of the following pairwise comparisons were statistically corrected.

Similarly to the previous methodological step, to rank the change (∆ in Figs. 8, 9) in the assessability of accident types, a group comparison was carried out for the various data recording technologies (T0, T1, T2). In order to diminish the probability of a Type I Error arising from the analysis of multiple hypothesis, the results of the following pairwise comparisons were statistically corrected.

The Kruskal–Wallis Test was applied in order to compare independent groups with a non-normal distribution [11].

$$H=\left(N-1\right)\frac{\sum_{i=1}^{g}{n}_{i}{({\overline{r} }_{i.}-\overline{r })}^{2}}{\sum_{i=1}^{g}\sum_{j=1}^{{n}_{i}}{({r}_{ij}-\overline{r })}^{2}}$$
(7)

where \({n}_{i}\), is the number of elements in the ith sample; \({r}_{ij}\), is the rank (among all elements) of the jth element in the ith sample; \(N\), is the number of elements across all groups; \({\overline{r} }_{i.}\), is the average rank of the ith sample; \(\overline{r }\), is the average rank of all the \({r}_{ij}\) in the unified sample; \(g\), is the number of samples.

In order to minimize the Type I Error, the Bonferroni Correction was used. The aim of the Bonferroni correction is the correct the unfavourable effect of the Type I error, which increases when comparing groups [2]. When multiple hypothesis tests are applied simultaneously, the probability of dropping the null hypothesis erroneously. In order to avoid this, the significance level (α) is lowered proportionally to the number of hypothesis tests (m). Consequently, the null hypothesis is dropped if for the value pi of the ith hypothesis test (\({p}_{i}<\frac{\alpha }{m}\)) proves to be true.

The applicability of the Bonferroni correction is supported by the following inequity [3], where m0 marks the number of true null hypotheses.

$$EF=P\left\{\bigcup_{i=1}^{{m}_{0}}\left({p}_{i}\le \frac{\alpha }{m}\right)\right\}\le \sum_{i=1}^{{m}_{0}}\left\{P\left({p}_{i}\le \frac{\alpha }{m}\right)\right\}={m}_{0}\frac{\alpha }{m}\le \alpha$$
(8)

3 Results and discussion

3.1 Evaluation of the current situation—traditional data recording

The following analysis compared the assessability of each accident type to that of the complementary set and the unified, total set. The rank averages of the examined sets were analyzed using the Mann–Whitney Test, as two independent groups were compared with non-normal distributions.

If the p value of the applied statistical test is lower than α = 0.05 (i.e. p < 0.05), the H1 hypothesis is accepted.

In the present case, Sample 1 is the set of the examined accident type, while Sample 2 is the complementary set, and the whole set of the accidents. The results of the statistical tests are given in Table 2.

Table 2 The assessability of accident types—T0—p values of the Mann–Whitney Test
Full size table

The values in Table 2 prove that the assessability of categories 5 and 7 is significantly lower than that of other accident types and also than that of the whole set of accidents for the T0 technique. Accidents occurring while lane changing or at crossings with traffic lights are considerably less assessable than other accident types.

The reason for this is that the data recorded after the collision involving lane change are usually insufficient for determining the route of the given vehicle prior to the crash. At the same time, for determining the liability for the collision, it is crucial to decide which driver started the given maneuver earlier in a way that the other driver could sense it.

As for accidents in crossings with traffic lights, responsibility crucially depends on the fact which driver entered the crossing when the traffic lights were green. However, the data recorded at the accident site are generally not sufficient for determining this.

3.2 Forecast—EDR technology

The aim of this step of the analysis was to estimate how much accident assessability would increase if EDR data were available. For this, the assessability of a given accident type using T0 and T1 data was compared to the assessability of the complementary set without T1 data, and the total set without T1 data, respectively.

Furthermore, it is also estimated how assessability will increase when T1 technology is spread. For this, the above comparisons are carried out, but the availability of T1 data is supposed for all sets (i.e. the given type, the complementary set, the total set of accidents).

In Table 3, Columns 1 and 3 show the p value for the comparison of the assessability of the given accident type set with T1 data and that of the complementary set (without and with T1 data, respectively). Columns 2 and 4 show the p value for the comparison of the assessability of the given accident type set with T1 data and that of the total set (without and with T1 data, respectively).

Table 3 The estimated assessability of accident types with T1 data—p values of the Mann–Whitney Test—values below the significance level are indicated with an asterisk (*)
Full size table

For Type 5 accidents, the inclusion of T1 data would improve assessability (cf. Columns 1 and 2 in Table 2). This result is in line with expectations. EDR data make it possible to reconstruct the route of vehicles prior to the collision. Consequently, the accident is more assessable.

However, as for Type 7 accidents, even data provided by the T1 technology prove insufficient to determine which vehicle entered the crossing while the traffic lights were green or red. Consequently, the assessability of such accidents is not enhanced.

Moreover, due to the general improvement in assessability owing to the EDR technology, the assessability of Type 5 (lane change) accidents is expected to stay below the general level of assessability (Columns 3 and 4 in Table 3). The same columns prove that the assessability level of Type 7 (traffic lights) accidents is not enhanced by the introduction of EDR technology.

3.3 Forecast—EDR+ technology

The following step in the analysis aimed to explore the extent to which the availability of T2data would further enhance the assessability of accidents. In this process, the assessability of a given accident type using data from T0and T2 systems was compared to the assessability of the complementary set without T2 data, and the total set without T2 data, respectively. Furthermore, it is also estimated how the assessment of accident types would improve if T2 data were widely available. In this latter test, the assessability of a given accident type is compared to that of the complementary set and the total set provided that T0 and T2 data are available for each element of the sets.

In Table 4, Columns 1 and 3 show the p value for the comparison of the assessability of the given accident type set with T2 data and that of the complementary set (without and with T2 data, respectively). Columns 2 and 4 show the p value for the comparison of the assessability of the given accident type set with T2 data and that of the total set (without and with T2 data, respectively).

Table 4 The estimated assessability of accident types with T2 data—p values of the Mann–Whitney Test—values below the significance level are indicated with an asterisk (*)
Full size table

For each examined accident type, the inclusion of T2 data would make assessability satisfactory compared to assessability of T0 methods (cf. Columns 1 and 2 in Table 2). However, as the assessability of all accident types increases if T2 data are available, the assessability of types 1 and 7 prove to be lower than the average assessability of the whole set (cf. Columns 3 and 4 in Table 2). The reason for this is that some accidents belonging to these types cannot be fully reconstructed even if T2 data were available for all participants.

3.4 Forecast—comparison of actual and future assessability by accident type

The next step of the analysis compared the assessability of a given accident type according to the three data category: (i) T0; (ii) T0 + T1; (iii) T0 + T1 + T2. The two matched samples with non-normal distribution were compared using the Wilcoxon Test.

The results of these statistical test are given in Table 5. If the baseline assessability is not significantly lower than estimated assessability if T1 or T2 data are available, the values are underlined.

Table 5 Assessability of different accident types according to the level of available data—p values of the Wilcoxon Test
Full size table

If data recording is expanded by the T1 technology, the assessability of accident Types 1 and 7 does not improve significantly compared to the baseline case.

As far as accident Type 1 is concerned, these collisions can be assessed to a satisfactory degree even based on T0 data. Consequently, for a significant increase in their assessability, extra information is required, which cannot be obtained by EDR systems. Such data could be collected by further individual data recording systems (e.g. a driver monitoring system).

Concerning accident Type 7, the EDR system does not record data that would significantly enhance the assessability of such collisions compared to the baseline case. However, owing to the GPS system, data collected by the EDR+ technology ensures the satisfactory assessability for this accident type.

3.5 Groupwise comparison of the assessability of accident types according to data recording technology

In the next step, the groupwise comparison of the assessability of accident types was carried out: (i) T0; (ii) T0 + T1; (iii) T0 + T1 + T2. This comparison made it possible to rank the assessability of different accident types for a given data recording technology. In this case, the groupwise comparison of independent samples with non-normal distribution was carried out, thus the Kruskal–Wallis Test was applied.

As a multiple hypothesis tests were carried out, in order to diminish the probability of Type I error, the significance level of the test was determined using the Bonferroni correction. If the p value of the statistical test is lower than the corrected α (p < αcorr), the H1 hypothesis is accepted. By comparing all possible sample pairs, it can be determined which pairs display a significant difference.

Table 6 shows the comparison of accident types according to data recording technique:: (1) T0; (2) T0 + T1; (3) T0 + T1 + T2.

Table 6 Groupwise comparison of the assessability of accident types for various data recording technologies—the p values of the Kruskal–Wallis Test—values below the significance level are indicated with an asterisk (*)
Full size table

For T0, there is stochastic dominance between samples 3–5 and 3–7; i.e. the values for elements randomly selected from the population characterized by the samples will be different. For T1, significant difference was found between four pairs: 3–5, 3–7, 5–6, and 6–7.

As for T2, for several pairs (2–3, 2–5, 2–6, 3–5, 3–6, 5–6) the value of p is not a number (NaN): for samples with equal elements, in the mathematical formula of the statistical test the denominator will be 0. Thus, no difference can be supposed between them. In practice, these accident types become fully assessable in the T2 case.

3.6 Comparison of sample pairs with stochastic dominance

Pairwise comparisons were carried out between those pairs that proved to display stochastic dominance with the aim to explore the relations between such pairs of accident types. If the p value of the statistical test is lower than α (p < α = 0.05), the H1 hypothesis is accepted.

Table 7 gives the results of the pairwise comparison of accident types: (i) T0; (ii) T0 + T1; (iii) T0 + T1 + T2.

Table 7 Pairwise comparison of the assessability of accident types for various data recording technologies—the p values of the Mann–Whitney Test
Full size table

The test results prove the following.

  • Incase T1, Type 3 (turning vehicle) is more assessable than Types 5 (lane change) and 7 (traffic lights).

  • In case T1, Types 3 (turning vehicle) and 6 (rear-end) would be more assessable than Types 5 (lane change) and 7 (traffic lights).

  • In case T2, most accident types would be better assessable that Type 1 (destabilization).

3.7 Groupwise comparison of the change in assessability of accident types according to data recording techniques

Groupwise comparison makes it possible to rank accident types according to the extent of change in assessability for a given data recording technology. Thus, independent samples with non-normal distribution were compared by the Kruskal–Wallis Test, combined with Bonferroni correction, in order to minimize the probability of Type I error during the multiple hypothesis test.

Table 8 shows the comparison of accident types according to data recording technique:: (i) T0; (ii) T0 + T1; (iii) T0 + T1 + T2.

Table 8 Groupwise comparison of the change in assessability of accident types according to data recording techniques—p values of the Kruskal–Wallis Test
Full size table

Data in Column 1 prove that if EDR data are introduced alongside data recorded by traditional techniques, changes in assessability do not significantly differ for the different accident types.

As for Column 2, if even EDR + data are available, the change in assessability compared to that if only traditionally recorded data are provided, significantly differs for some pairs: stochastic dominance is witnessed for pairs 1–2, 1–5, 1–6 and 3–5.

As regards the change in assessability between the cases with EDR and EDR + technology (Column 3), stochastic dominance is present in pairs 1–5, 3–5, and 5–6.

3.8 Comparison of sample pairs with stochastic dominance

For the next test, only those pairs were considered in which the previous test found stochastic dominance in order to reveal the relation between those accident types. The independent samples with non-normal distribution were compared using the Mann–Whitney Test. If the p value for the statistical test proves to be lower than α = 0.05, the H1 hypothesis is accepted.

Accident types are compared according to data recording technique in Table 9: Column 1 depicts the estimated change in assessability if T1techniques are introduced; Column 2 shows results for T0 + T1 + T2 compared to T0; while Column 3 illustrates the estimated change in assessability between cases with T1 and T2 data.

Table 9 Pairwise comparison of the change in assessability of accident types according to data recording techniques, compared to traditional and EDR technologies
Full size table

The following statements can be made based on the above results.

  • If EDR data are made available, the change in assessability of no accident type is significantly bigger than that of any other accident type.

  • If EDR + data are available, the change in assessability compared to traditional data recording techniques is significantly bigger for Types 2, 5 and 6 than for Type 1. Furthermore, the change for Type 5 is significantly bigger than for Type 3.

  • The change in assessability is significantly bigger for Type 5 than for Types 1, 3 and 6.

4 Conclusion

This study compared the levels of assessability of various road accident types with a series of statistical tests. The presented framework also makes it possible to examine how the assessability is expected to improve by the introduction of modern data recording technologies. Besides, our method contributes to the identification of the adequate data recording technology to achieve the required level of assessability.

The assessability of seven road accident types for forensic accident reconstruction was explored, using data from a collection of 124 real accidents in Hungary. Four assessability levels were established, and 7 accident types were set up. In addition to comparing the assessability of the various types, it was also estimated how assessability would change if data from new technologies (EDR; and EDR+, i.e. technology used by autonomous vehicles) were available.

The results prove that accident assessability considerably improves if EDR technology is applied in addition to traditional data recording technologies. If EDR+ data are also present, the assessability of each accident type would be satisfactory. As the assessability of each accident type grows owing to the EDR+ data, the assessability of accidents in Type 1 (roadway departure) and 7 (at traffic lights) does not reach the average. The reason for this is that some of the accidents in these sets could not be fully assessable (i.e. Level 4) even if EDR+ data were present.

Our evaluation proved that even if modern data recording technologies are applied, the assessability of accidents involving roadway departure (Type 1) and of accidents at traffic lights (Type 7) does not improve significantly compared to the case when only traditional data recording is used, as only information beyond data recorded by modern technologies could ensure full assessability (e.g. the physiological parameters of the drives for roadway departure).

In the next step, the groupwise comparison of the assessability of accident types was carried out according to various data recording techniques. The aim of this test was to rank the assessability of accident types for each data recording technology.

It is impossible to rank all accident types, as significant differences in assessability were found only for certain relations.

Finally, the groupwise comparison of the change in assessability for each accident type was completed to rank accident types according to the change in their assessability due to the introduction of modern technologies. This change did not prove to be significant for any type if only EDR technology is introduced. However, the inclusion of EDR+ data showed a lower change in assessability for roadway departure (Type 1) for other accident types. Furthermore, the assessability change for Type 5 (lane change) accidents in the EDR+ and EDR comparison is significantly higher than for several other accident types.

In conclusion, the above analyses proved that the assessability of some roadway accident types significantly differs from that of other types if only traditional authorized data recording technologies are applied (T0). This evaluative analysis supports the assumption that if modern data recording technologies (EDR, EDR+) widely spread (cases T1 and T2, respectively), the assessability of all accident types will increase. However, for some accident types full assessability cannot be reached. In order to enhance the assessability of these accident types, further data recording devices and techniques should be introduced. When choosing new methods, it is advisable to take into consideration how the application of the novel devices would affect the assessability of accidents. These effects vary according to device and accident type. The introduction of proper data recording technologies could enhance these methods’ level of usefulness to society.

In order to reach the appropriate assessability level of the examined seven types of accidents, in five cases, the traditional data recording technologies used today and, in some cases, the increasingly widespread EDR data recording technology are sufficient. At the same time, for Types 5 and 7, not even the EDR data recording technology can ensure adequate assessability; and in some cases, technologies beyond EDR+ would be necessary to achieve adequate assessability.

This research assumed that all the vehicles involved in a given accident were equipped with the same data recording technology. Thus it would be an important to examine those cases where the data recording technology of the vehicles is different since this difference may result in differences between the assessability of the accident forms.

For accidents involving highly automated vehicles, the use of EDR+ data recording technology is essential to ensure adequate assessability. Furthermore, with the further increase of the level of automation and the emergence and spread of autonomous systems, new types of accidents will result in the development and use of new analysis and data recording methods.