The objective of this second study is first to investigate whether a schema effect can be observed in residents’ spatial representations of their residing metropolitan area. Additionally, we also wanted to observe whether a schema effect would be found in the representation of a new city. As it would be newly learnt, the participants’ representations should match the learning material, and a schema effect should only be observed if the learning material was a transit map. We thus expected that the findings of this study will explain why the schema effect was observed in private transports users of Greater Paris, and whether it was indeed due to the transit map and exposure to it, or to heuristic biases in spatial representations.
We predicted that residents of a large metropolitan area will internalize specific metric violations present in their own transit network map. To test this hypothesis, participants from three different large European cities (Paris, London, and Berlin) were recruited. They were asked to produce maps of their region before having to learn one of the other city maps using a geographic or a schematic map and testing their spatial representation of this newly familiarised region. We expected participants to reproduce schematic distortions that were specific to their own city when they are tested on it. We also expected a congruency effect between the learning material and the test material for unknown cities. If newly learnt cities are encoded under their learning form (respectively geographic or schematic), known cities should be only represented under their schematic form because of the familiarity with their transport network.
1. Method
a. Participants
1103 participants were recruited to participate in an online study via a third-party recruiting platform (Dynata) and paid in exchange for their participation. Among this sample, 363 were French people living in or near Paris (\(\overline{M}\)age = 40.1, \(SD\)age = 12.4, women = 50.4%), 366 were English people living in or near London (\(\overline{M}\)age = 40.1, \(SD\)age = 11.8, women = 50.3%) and 374 were German people living in or near Berlin (\(\overline{M}\)age = 40.1, \(SD\)age = 11.8, women = 42.1%). All participants had lived in their respective city or city suburbs for more than four years and were frequent users of public transport (more than 4 trips a week before the COVID period). Informed consent was required to take part in the study.
b. Material
City and landmarks
To validate and extend the results of the first experiment, two new European metropolitan areas and their surrounding were selected, Greater London and Greater Berlin. These areas were selected because the general structure of their respective transport networks was similar to the Parisian one, but also because they have some minor differences that would help to distinguish specific effects linked to the use of transit maps (see Reference Maps Comparisons).
For London and Berlin, 10 well known landmarks were chosen, based on an analysis of their views on a tourism website (TripAdvisor) and responses to a questionnaire administered to residents of those metropolitan areas. For Paris we selected 10 of the 15 landmarks of the first studies based on their position with the objective of preserving central and peripheral locations of the region.
Three geographic maps were used for the learning task and as references for analyses. They were simplified maps of Paris, London and Berlin with their immediate respective regions. Although the actual areas of each depicted territory were different (area of the Paris region = 12012 km²; area of the London region = 1569 km²; area of the Berlin region = 891.8 km²), the three maps were presented at equal size depending on the resolution of the monitor used by the participant (64% of total full screen space, which corresponded to a 20×16 cm area for a standard 15-inch laptop screen). Each map showed the region’s border, major rivers, and the rail transport network symbolized by segments of different colours. The positions of the 10 landmarks were located on each map, along with a highlighted landmark which served as an absolute reference point (Notre-Dame for Paris, Picadilly Circus for London, and Brandenburg Tor for Berlin; see Fig. 7).
Reference Geographic Maps
Reference Schematic Maps
Three schematic maps have been used for the learning task and as references for analyses. These schematic diagrams were octolinear transit maps displaying the exact same layout as the official transport network map for each metropolitan area. They illustrate the same 10 landmarks as their geographic counterparts (with their positions adapted to the schematized shape of the network), as well as the same highlighted reference landmark whose position was now centred. Main rivers were displayed but schematized as well. Notably, no borders were displayed for reference schematic maps (see Fig. 8). Display size and actual covered areas were the same as geographic counterparts.
Test Layouts
Three different test layouts were designed for each city for the landmark-position-retrieval task. The first one reused the one developed in the first experiment. It was a blank layout except for the highlighted reference landmark, that was positioned according to the actual geographic reference frame (“blank sheet test material”). The second was the geographic test layout which only displayed the border of each region with their main rivers and the highlighted reference landmark, which was once again positioned according to the geographic reference frame (“geographic test material”). The last one was a blank layout similar to the white sheet test material, with the only difference that the highlighted reference landmark position was centred according to the schematic reference frame (“schematic test material”). The three layouts were used for the landmark-position-retrieval tasks performed on both known and newly learnt cities (see Fig. 9).
Questionnaire
Since all participants were required to be regular users of the public transport network in their area of residence, a questionnaire was created to verify that this requirement was adequately met. It consisted of 39 items among which were 32 visual analogue scales (yielding scores ranging from 0 to 100) that collected information on the frequency of use of public transport, personal vehicles, and route planning services in various situations. Two items were yes/no questions evaluating the current and past ownership of a personal vehicle. Four items were 10-point scales measuring the frequency of use of public transport over different periods, and one item was a 30-point scale recording the number of years lived in the participant’s home city or region. A few additional questions to this questionnaire were used to gather socio-demographic information.
c. Procedure
The entire protocol was created using the Gorilla experiment builder (Anwyl-Irvine et al., 2020). Participants accessed the study online on their personal computer by opening a link on any web browser. They first confirmed that they were between 18 and 65 years old, living in their city/near city for more than four years and that they used the public transport system of their city more than four times a week. They were then asked to read and sign a consent form before being presented with the instructions on study tasks.
The first phase of the study constituted of a landmark-position-retrieval task dealing with participants’ own city (known city). They were informed that they will have to position the landmark whose name will be indicated at the bottom of the screen, by using their mouse and answering as fast and precisely as possible. They were also informed that they will not be able to click twice and change their answer once they provide a position for a landmark. To make sure that they understood the task, an illustrated tutorial was provided to them with a step-by-step example. Then, the participants were randomly assigned a test layout material and performed the task by positioning each of the 10 landmarks on it. The landmark to be positioned was indicated by its name at the bottom of the screen. Landmarks’ order of presentation was randomised. No feedback was provided after each trial. Landmarks disappeared after being positioned so that each trial started on an empty test layout. Participants were not allowed to position the landmarks outside of the test layout.
Next, they moved on to the second phase and were informed that they will have to learn the locations of 10 landmarks situated in the region of another big European city (randomly assigned among the two cities in which they did not reside) (learnt city). This learning task was divided into two phases. In the first phase (passive learning), participants saw every landmark appear one after the other on a map. Each landmark appeared on its position indicated by a green dot with its name written above it and remained displayed for three seconds. This was repeated until the positions of the 10 landmarks were seen. The order of presentation of the landmarks was randomised. In the second phase (active learning), participants first saw the name of a landmark at the centre of their screen for three seconds. Then, the name was replaced by the same map used for the passive learning phase, and participants had to locate the landmark previously mentioned by clicking on a chosen position on the map with their mouse. Immediately after participants had given their answer, feedback was provided by displaying the actual position of the landmark for three seconds. This process was repeated three times for each of the 10 landmarks positions that were to be learned for the new city. Again, illustrated tutorials were provided to make sure that participants understood and performed the tasks as intended. The learning task was either performed using a geographic or a schematic reference map layout (participants were randomly assigned one of these reference materials). After all 10 landmarks’ positions of the novel city were learnt through passive and active learning, participants completed a second landmark-position-retrieval task for the newly learnt city. Test layout was once again randomised, and the instructions remained the same as during the first occurrence of this task.
Participants were then asked to complete the questionnaire before being provided with a debriefing explaining the aims of the study, communicating a contact address for questions and inquiries. Participants completed the study in less than half an hour (no specific time limit was imposed). Data was only recorded for participants who completed every step of the experiment.
d. Statistical Analyses
Analyses were performed using the R software 4.0.2 (R development Core Team, 2017). A bi-dimensional regression (Tobler, 1994; Friedman & Kohler, 2003) was first run to compare participants’ answers provided during the landmark-position-retrieval tasks to both the geographic and the schematic reference positions. This analysis is used to evaluate the degree of resemblance between a dependent and an independent two-dimensional configuration constituted by a set of coordinates. In this study, the dependent configuration is the whole set of positions given by participants during a single landmark-position-retrieval task (one for the known city and another for the learnt city). This set of coordinates estimated by participants can be reliably considered as a behavioural measure of certain metric properties of their spatial representations (Blades, 1990; Lhuillier et al., 2020). The bi-dimensional regression output provides several scores among which we chose the same measures as in experiment 1: bi-dimensional correlation coefficient, bi-dimensional scaling bias and bi-dimensional rotational bias (see Preliminary comparison between reference maps from experiment 1).
Bidimensional regressions were performed four times per participant: known and learnt city, each time with schematic and geographic reference configurations. We then performed several linear mixed-effects regressions (lme4 package version 1.1-8; Bates et al., 2015) using the bidimensional regression coefficients as dependent variables. Model were selected using the Akaike Information Criteria (AIC; Harrison et al., 2018) and χ² tests with a forward strategy model to avoid over-fitting.
This study used a between-subjects design. Two different models were considered for the known city and the newly learnt city, as the number of independent variables was not the same for each task. For the known city phase of the study, linear mixed-effect models included the reference map configuration (geographic or schematic), the test layout material (blank, geographic, or schematic) and the known city (Berlin, London, or Paris) as fixed effect factors with all possible pairwise interactions, with participants as a random-effect factor using random intercepts. For the second phase of the study (newly learnt city), the models included the known city (Berlin, London, or Paris), the learnt city (Berlin, London, or Paris), the reference map configuration (geographic or schematic), the reference learning material (geographic or schematic) and the test layout material (blank, geographic, or schematic) as fixed-effect factors. Only included pairwise interactions were between the reference map configuration, the reference learning material, and the test layout material. Once again, participants have been set as a random-effect factor with random intercepts. Each model was tested for abnormally influent observations using the Cook’s distance method (Cook & Weisberg, 1982). We used the standard cut-off value of Di < 4/n (Bollen & Jackman, 1990) to remove outlier observations before refitting the models. Pairwise comparisons within the mixed-effects models have been made using the Bonferroni-corrected probability of estimated group means differences (Searle, Speed & Miliken, 1980) using the emmeans package version 1.6.1 (https://cran.r-project.org/package=emmeans).
As in the first study, we only used scaling and rotation coefficients from the bidimensional regressions with the geographic reference map as independent variable (and not from the schematic map), to avoid circularity. Scaling bias values above one then indicate an expansion of inter-landmark distances on participants’ sketch maps and values below one indicate a compression of the inter-landmark distances (compared to the geographic map). Positive values of the rotational bias indicate a general clockwise rotation of inter-landmark angles on the sketch maps and negative values a general anticlockwise rotation (compared to the geographic map).
We used a linear regression for each of those metrics (scale and rotation), with the same fixed effects as the above-mentioned mixed effect regression – excluding reference map, as only output from the regression with the geographic reference map were taken. Each model was tested for abnormally influent observations using the Cook’s distance method (Cook & Weisberg, 1982). We used the standard cut-off value of Di < 4/n (Bollen & Jackman, 1990) to remove outlier observations before refitting the models.
Finally, we used Wilcoxon signed rank tests to test whether the scale biases were different from one (i.e., different from the geographic scale) and to the respective schematic scale (see Table 3) for each group. Similarly, we used Wilcoxon signed rank tests to test whether the rotational biases were different from zero (i.e., different from the geographic scale) and to the respective schematic scale (see Table 3) for each group.
2. Results
Preliminary comparison between reference maps
As in study 1, we performed a bidimensional regression on the two reference map configurations to assess the extent to which schematic maps are distorted compared to geographic maps. This analysis was conducted for all three metropolitan areas. The schematic map configuration was set as the dependent variable, and geographic map configuration as the independent variable. Results are shown in Table 3. This analysis allowed us to isolate three main characteristics of the cartographic material used in this study, each of which is specific to the schematic representation of a single city of interest. First, although all schematic maps are highly correlated with the geographic map, the Berlin schematic configuration stands out as the most similar to the original material as indicated by a higher correlation coefficient. Second, although all schematic maps globally expand distances compared to the geographic material, the London schematic map stands out as the most expanded representation, as indicated by the scaling bias score. Third, only the Paris schematic map shows a substantial rotation corresponding to 8.83 degrees clockwise, as indicated by the rotational bias score.
Table 3
Bidimensional Regression measures for the comparisons between each city’s schematic reference map and geographic reference map
|
Bidimensional Regression Scores
|
|
Correlation coefficient
|
Scaling bias
|
Rotational bias
|
Berlin
|
0.977
|
1.43
|
-0.26
|
London
|
0.944
|
1.52
|
0.29
|
Paris
|
0.943
|
1.47
|
8.83
|
a. Known Cities
Figure 10 Veridical (left) and plotted (right) landmarks positions by Parisians test on a blank layout (Top), a geographic layout (Middle) or a schematic layout (Bottom). See supplementary figures for other cities. The black line outlines the region (Ile-de-France), and coloured lines represent train and metro lines. Left: Black dots show actual landmark positions, blue dots show average plotted positions, and orange arrows show the difference between the locations of each landmark. Right: Anamorphic map of the region after geometric distortion according to participant answers.
Correlation coefficient
All participants’ correlation coefficients were significantly higher for the schematic reference (Mschema = 0.596, SD = 0.263) than for the geographic reference map (Mgeo = 0.575, SD = 0.26; β = 0.013, SE = 0.004, t = 3.62, p. < .001). This indicates a higher similarity towards the transit map, i.e., a schema effect. City also predicted coefficient, with Londoners obtaining significantly worse performances than Berliners (p. = .043). However, Parisians did not significantly differ from Berliners (β = -0.042, SE = 0.034, t = -1.362, p. = .173). The reference map also interacted with the layout, with the higher similarity to the transit map decreasing if the test layout was geographic (β = -0.015, SE = 0.005, t = -3.651, p. <.001; see. Figure 11). Finally, reference map interacted with city (β = 0.036, SE = 0.004, t = 8.883, p. <.001): whereas the schema effect exists for all cities, its amplitude is significantly larger for Paris than other cities (see. Figure 12). The model’s marginal R2 indicates that fixed effects explain 2.5% of the variance in participants’ responses.
Scaling bias
The geographic test layout material yielded maps on a smaller scale than the schematic test layout material (β = .19, SE = .06, t = 3.4, p < .001; Fig. 13), but not different from maps done on a blank layout material (β = .11, SE = .06, t = 1.9, p = .05). A significant main effect of the city was also found, indicating that Londoners compressed distances compared to Berliners (β = .13, SE = .06, t = 2.3, p < .05; Fig. 13) and Parisians (β = .21, SE = .06, t = 3.7, p < .001; Fig. 13). Individual Wilcoxon tests showed the scale was below the respective schematic scale for every group, and below the geographic scale (i.e., 1 in Fig. 13) for every group except Parisians with a blank layout (V = 2632, p = .09).
Rotational bias
Figure 14 shows different rotation patterns were found for each city. We thus performed separate linear regressions for each city. Londoners rotated their map more clockwise with a blank test layout material (β = 10.2, SE = 3.7, t = 2.8, p < .01) or a schematic test layout material (β = 8.7, SE = 3.5, t = 2.5, p < .05) than with a geographic test layout material. However, the opposite was found in Parisians, with a more clockwise rotated map with a geographic test layout material than with a blank (β = -9.3, SE = 4.0, t = -2.3, p < .05) or schematic (β = -11.3, SE = 4, t = -2.9, p < .01) one. Berliners rotated their maps more clockwise with a geographic test layout material than with a blank (β = -7.3, SE = 3.4, t = -2.1, p < .05) but not differently to with a schematic one. Individual Wilcoxon tests showed that for Parisians, only rotations from a test on the geographic layout were different from the geographic orientation (V = 4747, p < .05) but not different from the schematic orientation. Berliners tested on the geographic layout had orientations significantly different from both geographic (V = 5228, p < .01) and schematic orientations (V = 5196, p < .05), but tested on a schematic layout they were only different to geographic orientation (V = 4747, p < .05), and tested on a blank layout they were different to neither. Londoners tested on the schematic or blank layout had a clockwise rotation relative to both schematic and geographic orientation (all ps < .05), but tested on a geographical layout there were different to neither.
b. Learnt Cities
Similar analyses were performed on the sketch maps produced from a newly learned city.
Correlation coefficient
A significant coefficient was found for the newly learnt city of London (β = 0.049, SE = 0.022, t = 2.264, p. = .024, Fig. 15), suggesting that learning the positions of London’s landmarks led to better results than other cities, independent of the learning material or the test material.
A significant interaction effect was observed between the reference map configuration and the schematic learning reference material (β = 0.052, SE = 0.003, t = 15.691, p.< .001). Pairwise comparisons revealed the existence of a congruency effect (see Fig. 16) - learning a schematic reference map led to higher correlation with the schematic reference configuration (t = 14.724, p.< .001), whereas learning a geographic reference map led to higher correlation with the geographic reference configuration (t = -7.321, p.< .001). Another significant coefficient was found regarding the interaction between the test layout material and the reference map configuration (β = .015, SE = 0.004, t = -2.519, p. = .012). Pairwise comparisons revealed that correlation with the schematic reference map decreased if the retrieval task was performed on a geographic test layout (t = -2.255, p. = .024), but this effect was not observed for the blank test layout (t = 0.397, p. = .691) or the schematic test layout (t = -0.119, p. = .905). The model’s marginal R2 indicates that fixed effects explained 1.6% of the variance in participants’ responses.
Scaling bias
The geographic test layout material yielded maps on a smaller scale than the schematic test layout material (β = .08, SE = .04, t = 2.1, p < .05; Fig. 17) and with a blank test layout material (β = .14, SE = .04, t = 3.6, p < .001). A significant main effect of the learnt city was also found, indicating that Berlin was more compressed than London (β = − .13, SE = .03, t = -4.0, p < .001). No main effect of learning material was found, but it significantly interacted with test material: the scale was larger for participants who learned on a schematic material, if they were then tested on the blank layout (β = .22, SE = .06, t = 3.9, p < .001) or the schematic layout (β = .29, SE = .06, t = 5.2, p < .001).
Individual Wilcoxon tests showed the scale was below the respective schematic scale for every group, and below the geographic scale (i.e., 1 in Fig. 17) for every group except all those who learned on a schematic layout and were then tested on a blank layout (for all learnt cities), or a schematic layout (only for London and Paris as learnt city).
Rotational bias
As in study 1, we did a separate regression for each learnt city to better capture the variability of patterns. However, no main or interaction effect was significant. Individual Wilcoxon tests showed that no rotation was significantly different from either the schematic and geographic orientation for London or Berlin as learnt city. However quite oddly, people who learnt Paris on a schematic layout were significantly different from both geographic and schematic rotations, as they rotated anti-clockwise (all ps < .05, Fig. 18). On the contrary, Paris sketch maps were rotated clockwise if they were learnt and tested on geographic layouts (V = 779, p < .05).
3. Discussion
The aim of this second study was to further investigate the origin of the schema effect observed in Study 1. More specifically, we asked the following question: does the schema effect arise from a set of cognitive representational heuristics which function to regularize and simplify the processing of spatial information, or is it a consequence of familiarity with schematic symbolic representations such as transit maps? Taken together, our findings point towards the second interpretation of the schema effect: familiarity with the schematic representation seems to have a direct impact on the metric properties of the spatial representation in memory. This conclusion comes from the observation of two effects that we describe below: 1) when asked to freely recall the spatial positions of their city component elements, residents reproduce metric biases specific to their own city's transportation schema 2) when they learn a map of an unfamiliar city, their representation is congruent with the learned format (either geographic or schematic) and is hermetic to the metric biases specific to their city of residence.
The main results of the first phase of Study 2 showed that participants’ spatial representations were more correlated with the schematic transit map of their region than with the geographic map, which is a replication of the schema effect observed in Study 1. More precisely, Study 2 demonstrated that although this schema effect exists for known cities regardless of the material used for retrieval, its amplitude decreases when participants are tested on a geographic layout. These results suggest that participants’ spatial representation of their resident city is more dependent on the schematic layout coordinates (with which they are familiar) rather than the actual geographic coordinates of locations. This holds true even when actual geographic borders are provided during retrieval, although the magnitude of the schema effect decreases in this condition.
In line with Study 1, we also found that the amplitude of the schema effect was significantly larger for Parisians than for other cities’ residents, which is in line with Study 1 findings. This could explain the absence of difference between private and public transports users observed in experiment 1. As noted by Roberts et al. (2013), Paris has a remarkably high number of stations (around 380) within its transportation network compared to London or Berlin, which makes it a uniquely interconnected city (Roberts, 2016). As a result of this, the network of Paris’ stations could be used by its residents as an effective grid of landmarks to underpin spatial knowledge, thus making the transit map even more valuable than its geographical counterpart. Another explanation that could be investigated is the fact that the Parisian schematic map covers a larger area than other cities’ maps. This could force Parisians to a higher level of regularization in their representation, to compensate for the larger amplitude of the territory to be encoded (Milgram & Jodelet, 1976).
In addition, Berliners (whose schematic map correlates best with the geographic map) show overall higher mean correlation scores than Londoners. Although the difference with Parisians was not significant, the schema effect is also greatly reduced in Berliners compared to Parisians.
However, we also observed an unexpected result regarding the specific distortion associated with the London schematic map: although this transit map was substantially more expanded than the other two cities’ maps, Londoners’ sketch maps were significantly smaller than the other two cities. This might actually be due to London’s larger expansion on the transit map, as Londoners might have wanted to leave more space for distant landmarks. In any case, sketch maps were smaller than both geographic and schematic scales. This suggest that participants tend to leave a blank margin on the side, perhaps because of some border effect (Tversky, 1981), rather than reliance on either reference map. This could indeed be the case, as sketch maps were also smaller if the test layout was geographic, i.e., if the region’s borders were shown on the layout.
To test whether the schema effect is better explained by participants’ familiarity with the schematic transit map rather than cognitive regularization processes, a learning task of an unknown city was used in a second phase. As expected, correlation scores revealed that participants’ representations more closely resembled the reference layout (i.e., schematic, or geographic) that was presented during the learning phase than the other layout not seen during this phase.
In addition, we observed a congruency effect between learning material and retrieval material: learning a new region using a schematic reference layout leads to a higher correlation with the schematic reference map if participants are later tested on a schematic layout rather than a geographic or a blank layout. Conversely, participants who learned using a geographic reference map show higher correlation scores with the geographic reference map when tested on a geographic layout rather than a schematic or a blank layout. Taken together, the results of this study globally support the hypothesis that schematization biases arise from familiarity with cultural representations and symbolic tools rather than from heuristic regularization processes.
For scaling bias, the pattern was partly similar: participants who learned the new city on a schematic layout drew bigger sketch maps, but only if they were tested on a blank or schematic layout. The borders on the geographic test layout thus seems to have led all participants to draw smaller maps.
It is important to note that the differences in density between central and peripheral areas have not been correctly reflected in this study. Network (and population) density tends to be lower the more one moves away from the city centre. As such, there are more 'empty spaces' in the peripheries when we look at a geographic map of the public transport network. Consequently, transit maps usually remove these "empty spaces" for better readability and usability - thus bringing the peripheries closer to the centre. Future studies should aim to clarify these scaling effects by taking into account localized distortions between central and peripheral metropolitan areas, since disparities in station density should be accompanied by directional biases in spatial memory.
Taken as a whole, the findings of these two experimental studies contribute to clarify the observations made by Vertesi (2008). We applied an interdisciplinary perspective combining cognitive psychology and geography, through the adoption of an experimental methodology and bidimensional spatial analyses. This quantitative approach allowed for the testing of precise hypotheses dealing with specific distortions of mental spatial representations, associated with the use of different cartographic supports. From a theoretical standpoint, we provide yet another argument towards the integrative nature of spatial representations (Tversky, 2003). Indeed, we show that spatial representations of big cities are encoded from multiple sources, one of which is the schematic transit map with which residents seem to be familiar. The influence of transit maps’ distortions on spatial memory metric properties illustrates how mental representations are also functionally oriented, as deviations from actual topography could emerge from implicit time-cost route planning heuristics (Montello, 2009). This is coherent with the idea that spatial representations are non-isomorphic by nature because it helps supporting novel route finding and shortcut generation (Warren et al., 2017) within a complex network.
From a practical and an applied standpoint, the findings presented in this paper provide an argument towards the necessity for mass transit service providers to take into careful consideration the impact of the symbolic representations of their transportation networks. Indeed, in addition to communicating information regarding the workings of the network, transit maps also shape residents’ mental representations of the environment in the long term. Complementing this, previous research has illustrated that travellers tend to interpret information on the transit map as corresponding to a reality in the physical world – a complex codification of a transfer station will lead travellers to perceive the station as a complex one to traverse during a transfer in the physical word (Guo, 2011; Grison et al., 2022). In a bid to harmonise passenger flow across the transportation network, mass transit service providers may also choose to harness the impacts of transit maps on residents’ mental representations of the city. For instance, it may be possible to promote route choices to residents based on where they live in the city and how they have come to internalise the spatial relations between different places in the city. However, in order to achieve this at a highly efficient degree, as cited previously it will be useful to look more closely into more specific and localised effects of the transit map on residents’ mental representations.