Heterogeneous correlations between hippocampus volume and cognitive map accuracy among healthy young adults

Abstract

Marked individual differences in the ability to mentally map our environment are pronounced not only among people of different ages or clinical conditions, but also within healthy young adults. Previous studies have shown that hippocampus size positively correlated with spatial navigation ability in healthy young adults, navigation experts, and patients with hippocampus lesions. However, a recent pre-registered study (Weisberg, Newcombe, & Chatterjee, 2019) with a large sample size (n = 90) did not observe this correlation in healthy young adults. Motivated by evidence that self-report sense of direction (SOD) could have a profound impact on how individuals utilize environmental cues, and that different navigation strategies could have opposite impacts on wayfinding performance in individuals with different cognitive map formation (CMF) abilities, we reanalyzed the publicly available dataset from Weisberg et al’s study. We tested the influence of participants’ SOD and CMF abilities on hippocampal volume-performance relationships. We find evidence that the non-significant correlation could envelop heterogeneous correlations among subgroups of individuals: the correlation between the right posterior hippocampal volume and spatial learning performance is significantly higher among individuals with high spatial ability than individuals with low spatial ability. This pattern of performance was observed for both SOD and CMF moderations of the relationship between hippocampal volume and spatial learning. While our re-analyses are fundamentally exploratory in nature, the new results imply that the relationship between hippocampal volume and spatial learning performance might be more complicated than previously thought.

1. Introduction

Spatial navigation is one of the most fundamental functions in our daily life, and this ability varies considerably among individuals (Wolbers & Hegarty, 2010). On the expert end, London taxi drivers have to pass a series of very stringent tests for employment, which includes memorizing the city’s labyrinthine-like 25,000 streets. On the opposite end, some people have a very difficult time orienting themselves even in extremely familiar surroundings, despite the absence of any acquired brain damage or neurological disorder (Iaria & Burles, 2016). What could possibly contribute to these profound individual differences?

One contributing factor to such individual differences may be the size of hippocampus. Maguire et al. (2000; 2003) showed that London taxi drivers, who are required to deliver passengers to their destinations flexibly and by the shortest route, had larger right posterior hippocampi than healthy-controls and London bus drivers. Interestingly, the taxi drivers’ anterior hippocampi were smaller than control participants (Maguire et al., 2000). Myriad studies have since shown in healthy young college students that hippocampal volume is correlated with spatial learning performance (e.g., Brown, Whiteman, Aselcioglu, & Stern, 2014; Chrastil, Sherrill, Aselcioglu, Hasselmo, & Stern, 2017; Schinazi, Nardi, Newcombe, Shipley, & Epstein, 2013; Sherrill, Chrastil, Aselcioglu, Hasselmo, & Stern, 2018). Although the sample size in some of those studies were small (e.g., Maguire et al., 2000; 2003), warranting caution with the correlations, such data suggest that, across studies in which spatial knowledge was measured in different manners, larger hippocampi often predict better spatial navigation performance.

A recent study, however, challenged this conclusion, at least among healthy young adults. Weisberg et al. (2019) recruited ninety participants for Magnetic Resonance Imaging (MRI) and conducted carefully pre-registered tests commonly taken to measure cognitive map accuracy (pointing to buildings which cannot be seen from the current bearing, placing model buildings onto a map) in a novel environment. Contrary to the positive correlations reported in previous studies, Weisberg et al. (2019) consistently found no evidence that hippocampus volume (anterior or posterior, left or right hemisphere), or volume of related medial temporal structures (parahippocampus, entorhinal cortex), predicted metrics of spatial memory (pointing and model building tasks).

One possible explanation for the difference between Weisberg et al. (2019) and previous studies is that others have observed but not reported non-significant correlations (the “file drawer problem”) or publication bias (Rosenthal, 1979): It is difficult for researchers to publish studies with null results, especially when the sample size is small and the study is not pre-registered. On the other hand, carefully-conducted, pre-registered research with a large sample size such as Weisberg et al. (2019) enable the field to overcome this bias and to show a more comprehensive picture. It may be that the volume-performance relationship is quite indirect and is mediated by other factors, but the revelation of such bridging mechanisms has been held back by the file drawer problem or publication bias. Given the potential theoretical impact from Weisberg et al. (2019) due to its pre-registered nature and large sample size, we believe that it is imperative that the field now explore different possibilities for why results differ across tasks and research groups (important possibilities raised by Weisberg and colleagues include the idea that observed navigational ability is an emergent property of many factors, and the structural properties of the hippocampus and behavior may therefore have a complex relationship). We believe that one good way to explore the differences is to re-analyze Weisberg et al.’s (2019) publicly available data (https://osf.io/ea99d/).

Our targeted, exploratory re-analysis is motivated by two sets of findings (He, McNamara, & Brown, 2019; Weisberg & Newcombe, 2015), each of which suggests that individuals can be divided into meaningful navigator sub-groups (e.g., their sense of direction, ability to form an integrated cognitive map) which can interact with other factors (e.g., environmental cue availability, spatial navigation strategy) to influence observed spatial learning performance.

Firstly, our recent study (He, McNamara, & Brown, 2019) showed that self-report sense of direction (SOD) modulated the efficacy of spatial cue intervention. In this study, participants freely explored in a virtual environment where the environmental barriers were either opaque or translucent. Participants’ wayfinding efficiency and cognitive map accuracy were measured in an opaque environment after the free exploration. We hypothesized that the translucency treatment could enable participants to observe spatial relationships between locations directly and therefore their spatial learning should be facilitated compared to the opaque conditions where the spatial relationships between out of sight locations must be inferred. We found that only the individuals with high SOD could benefit from the translucency treatment and outperformed their high SOD counterparts in the opaque conditions. This pattern of results was not observed for the individuals with low SOD. Our results suggest that adding informative environmental cues does not necessarily facilitate spatial learning, per se, but only individuals with high SOD utilize them to form a more accurate cognitive map.

Secondly, Weisberg and Newcombe (2015) showed that the same spatial strategy had opposite effects on cognitive map tests for individuals with good and bad cognitive map formation (CMF) ability. The authors measured individual’s spatial strategy preference with a dual-solution paradigm (Marchette, Bakker, & Shelton, 2011) and CMF abilities with their virtual Silcton task (Weisberg, Schinazi, Newcombe, Shipley, & Epstein, 2014). The authors divided participants into three groups based on their virtual Silcton task performance (from high to low spatial ability: Integrators, non-integrators, imprecise navigators), and found no correlation between groups and strategy preference, suggesting that CMF ability and spatial strategy preference could be independent. More importantly, Weisberg and Newcombe (2015) reported that the more the Integrators preferred the “place strategy”, which is a strategy focused on learning the configuration of the environment, the more goals they found in the dual-solution shortcut test. Strikingly, the more the imprecise navigators preferred the “place strategy”, the fewer goals they found. In other words, good navigators do not necessarily prefer a place strategy, and a place strategy may be less suitable for imprecise navigators.

Taking the results from the two studies together, it seems that good and bad navigators, whether they are categorized subjectively (He et al., 2019) or objectively (Weisberg and Newcombe, 2015), utilize environmental cues or cognitive resources very differently: when considering observed spatial ability as an emergent property, “good navigators” may be those who do not simply have a big hippocampus, but use cues, strategies, and hippocampal memory resources flexibly to facilitate CMF. “Bad navigators” may still have positive traits (e.g., large memory capacity), but be individuals who do not take full advantage of trait and environmental resources (He et al., 2019) or even misallocate them (e.g., adopt a sub-optimal strategy¹; Weisberg and Newcombe, 2015).

We posit that hippocampal volume is a general relational memory resource (Eichenbaum, 2017), and we conjecture that good and bad navigators both utilize their hippocampi when encoding a navigational task such as Silcton. Thus, following He et al. (2019) and Weisberg and Newcombe (2015), although the hippocampus is a resource critical in cognitive map formation (O’Keefe & Nadel, 1978), it is possible that good navigators are those who use the mnemonic resources provided by the hippocampus to encode episodes in the environment in a manner optimal for subsequent tests of CMF. Bad navigators, on the other hand, may not optimally allocate mnemonic resources for tests of CMF. It is with this alignment that we speculate performance would be expected to scale positively with more resources. As reviewed in Goodroe, Starnes and Brown (2018), the hippocampus can support encoding from a variety of reference frames, and the information that gets encoded in declarative memory need not be optimal for a participant’s strategies (Weisberg and Newcombe, 2015) or the test they are later given.

To test whether SOD and tendency to integrate routes can moderate the relationship between hippocampal volume and spatial learning performance, we used the Santa Barbara Sense of Direction scale (Hegarty, Richardson, Montello, Lovelace, & Subbiah, 2002) and objective indicators of Integration ability/tendency (virtual Silcton), as moderator variables. To limit the scope of our exploratory analysis and reduce Type I error rate, we focused on the anterior and posterior subregions of the right hippocampus as our regions of interest (ROI). This was based on prior studies showing that the volume of these regions correlates with spatial learning (Brown et al., 2014; Chrastil et al., 2017; Maguire et al., 2000; Maguire et al., 2003; Sherrill et al., 2018).

For our dependent measure of cognitive map accuracy we focused on Weisberg et al.’s (2019)’s model building task. Both the model building task and the pointing task have been considered good measures to represent cognitive map accuracy (He, McNamara, Bodenheimer, & Klippel, 2019; Ishikawa & Montello, 2006; Weisberg et al., 2014). One may argue that the between-route pointing task, which requires participants to integrate locations across geographical regions (see Materials and methods), is an ideal representation of cognitive map accuracy. This argument can be made on the fact that the between-route pointing resembles requiring participants to find a shortcut, which has been considered as the hallmark of a cognitive map (Tolman, 1948). However, recent computer simulations demonstrate that a cognitive map is not necessary for finding a shortcut (Cruse & Wehner, 2011). On the other hand, we consider that if participants know the locations of buildings from an allocentric perspective (model building task), then they should be able to derive shortcuts from this allocentric representation. Importantly for us, while the pointing judgments are made from an egocentric reference frame, the model building task requires participants to convert the spatial knowledge acquired from an egocentric perspective during encoding to an allocentric perspective at test, which is a key component in a cognitive map (Siegel & White, 1975) and made this a particularly compelling measure that they developed.

Based on the findings of our previous study (He et al., 2019) and Weisberg et al.’s (2015), we predicted that the correlation between hippocampal volume and the model-building task performance was positive among individuals with high SOD or integrators, whereas this correlation would be non-significant (He et al., 2019) or even negative (Weisberg et al., 2015) among individuals with low SOD or imprecise navigators. Finally, these correlations should be significantly different between high SOD and low SOD individuals, and/or different between integrators and imprecise navigators. Our re-analysis largely supported our predictions of the SOD division.

2. Materials and methods

Full descriptions of this section can be found in Weisberg et al. (2019). Here, we focus on the relevant materials and methods for our reanalysis. No part of the reanalysis reported in this paper was pre-registered prior to the research being conducted. Pre-registration of study procedures is not applicable as the current study involved reanalysis only and did not include newly-acquired data.

Ninety participants (54 women) were recruited to undergo an MRI scan and completed a battery of tests. The volume estimates of the right posterior hippocampus were extracted manually using FreeSurfer’s automated hippocampal parcellation. Before the spatial learning experiment, participants first completed a series of behavioral and self-report measures. Critical for our re-analysis was the Santa Barbara Sense of Direction scale (SBSOD; Hegarty et al., 2002). This self-report measure of navigation ability consists of fifteen 7-point Likert-scale items such as “I am very good at giving directions,” and “I very easily get lost in a new city.” The average score for each participant has been shown to correlate highly with performance on behavioral navigation tasks in real and virtual environments (Hegarty et al., 2002).

After completing these measures, participants then performed the Virtual Silcton task (Weisberg et al., 2014). Participants used mouse and keyboard to navigate in a virtual environment. During navigation, participants learned the names and locations of eight buildings in two separate areas (four buildings each) of the same virtual environment in constrained paths. Participants were unaware of the nature of subsequent cognitive map testing, such that their encoding strategy and reference frame were not constrained.

After learning, participants’ spatial knowledge of the environment was probed by two tasks. For an onsite pointing task, participants were first teleported to one of the eight buildings, and pointed to all other buildings that they learned. Because the to-be-pointed buildings could be within the same route or across routes relative to the building where the participants were at, Weisberg et al. (2019) divided the pointing trials into within-route and between-route ones. The second test was a model-building task wherein a map was displayed on a computer screen from an aerial perspective. Participants placed the eight learned buildings to the locations where participants believed they would be located in the map, without regard to the orientation of the buildings or to the map. Performance in the model building task was measured in two ways: 1) Model building Total. This was a bidimensional correlation coefficient for all eight buildings. 2) Model building Within-route. It reflects the average of bidimensional correlation coefficients for the four buildings within each route. Although both the Model building Total and Model building Within-route tax the cognitive map given the allocentric configural nature of the test, we consider that Model building Total may better reflect integrated cognitive map memory across the route experiences.

3. Results

3.1. SOD moderated the relationship between right posterior hippocampal volume and the model building task performance

Based on our prior work (He et al, 2019), we first examined whether SOD moderated the relationship between hippocampal volume and cognitive map accuracy. Because categorizing participants’ SOD into groups introduces arbitrary boundaries in the data, we treated participants’ SOD as a continuous variable in a moderation analysis. Whole brain volume, ROI volume (anterior or posterior right hippocampus) and grouping measure (SOD or virtual Silcton grouping) were entered in the first step of the regression model and the interaction term (ROI volume X grouping measure) was entered in the second step. A significant interaction term indicates a moderation. For brevity, we only report the statistics of the interaction term here, and the descriptions of the full model can be found in the Supplemental Materials. All the patterns of the results reported here remained the same when whole brain volume was not controlled for. Given that our specific moderator and correlational analyses were directly motivated by previous studies (He et al., 2019; Weisberg et al., 2015), we did not correct the α value ( = 0.05) for multiple comparisons, but readers should be aware of the potential for inflated Type I error rate in such cases.

For model building total, the interaction between SOD and the right posterior hippocampal volume was marginally significant (t(89) = 1.887, p = 0.063). No such trend was observed in the interaction between SOD and the right anterior hippocampal volume (t(89) = 0.787, p = 0.378).

For model building within-route, the interaction between SOD and the right posterior hippocampal volume was significant (t(89) = 2.167, p = 0.033). No such trend was observed in the interaction between SOD and the right anterior hippocampal volume (t(89) = −0.712, p = 0.864).

In sum, SOD either marginally significantly or significantly moderated the relationship between right posterior hippocampal volume and the model building task performance. To visualize how such relationship changed across participants with different SOD, we divided participants into high, medium and low SOD groups (approximately 30 participants in each group) and plotted them in Figure 1. For the model building total, we found that the correlation was significantly negative in the low SOD group (r = −0.35, p = 0.048; Fig.1). By contrast, the correlation was significantly positive in the high SOD group (r = 0.39, p = 0.034; Fig.1), and the correlation in the high SOD group was significantly more positive than the low SOD group (Z = 2.91, p = 0.004). For the model building within-route, we found that the correlation was significantly negative in the low SOD group (r = −0.41, p = 0.018; Fig.1). In addition, the correlation in the high SOD group was significantly more positive than the low SOD group (Z = 2.39, p = 0.017).

Fig. 1 –

Correlations between model building task performance and right posterior hippocampal volume across low, medium and high SOD individuals. Upper panel: model building total; lower panel: model building within-route. *p < 0.05, *** p < 0.005

3.2. Silcton grouping moderated the relationship between right posterior hippocampal volume and the model building task performance

Based on the pointing task performance, Weisberg et al. (2019) divided participants into three groups: Integrator, non-integrator and imprecise navigator. Integrators were the individuals who performed well in both the within- and between-route pointing trials. Non-integrators were the individuals who performed well in the within-route pointing trials but poorly in the between-route pointing trials. Imprecise navigators were the individuals who performed poorly in both pointing trials. The moderation analyses here were similar to the ones reported above but with Weisberg et al.’s (2019) grouping label as moderator variable.

For model building total, the interaction between Silcton grouping and the right posterior hippocampal volume was not significant (t(89) = 0.548, p = 0.585). A similar pattern was observed in the interaction between Silcton grouping and the right anterior hippocampal volume (t(89) = 0.099, p = 0.921).

For model building within-route, the interaction between Silcton grouping and the right posterior hippocampal volume was significant (t(89) = 2.272, p = 0.026). No such trend was observed in the interaction between Silcton grouping and the right anterior hippocampal volume (t(89) = −0.531, p = 0.597).

In sum, Silcton grouping only significantly moderated the relationship between right posterior hippocampal volume and model building within-route. This contrasted with SOD and our predictions, suggesting that SOD may measure a partially distinct (and for this specific experiment, seemingly a qualitatively more powerful) trait moderator. To better visualize how this relationship changed across participants in different groups and to compare with our SOD analysis, we plotted the correlations between right posterior hippocampal volume and model building task performance in Figure 2. Unlike with SOD moderation, for the model building total we did not find any significant correlations in any subgroup or any significant differences between any subgroup. For the model building within-route measure, we found that although the correlations within each group was not significantly different from zero (Fig.2), the correlation in the integrator group was significantly more positive than the imprecise (Z = 2.07, p = 0.038) and the non-integrator groups (Z = 2.21, p = 0.027).

Fig. 2 –

Correlations between model building task performance and right posterior hippocampal volume across imprecise navigators, non-integrators and integrators. Upper panel: model building total; lower panel: model building within-route. *p < 0.05.

3.3. SOD did not moderate the relationship between right hippocampal volume and the overall pointing task performance

As discussed in the Introduction, we focused on the model building task as a measure of cognitive map accuracy to limit the number of tests and reduce Type I error rate. However, to better compare with Weisberg et al.’s (2019) principal pre-registered analysis (correlation between right hippocampal volume and the overall pointing task performance), it was of interest to also run their analysis with SOD as a moderator variable. For this analysis, we could not use the Silcton grouping as a moderator variable because the pointing task performance was also used for Silcton grouping. Unlike the Model building total data, we did not find that SOD moderated the volume-pointing performance relationship (t(89) = −1.246, p = 0.216).

4. Discussion

Weisberg et al. (2019) reported that hippocampal volume was not correlated with spatial learning performance among heathy young adults in Silcton. This differed from previous findings also in healthy young adults (Brown et al., 2014; Chrastil et al., 2017; Schinazi et al., 2013; Sherrill et al., 2018). We hypothesized that the non-significant results from Weisberg et al. (2019) could reflect heterogenous correlations among participants with different spatial ability. By conducting moderation analyses with participants’ SOD as moderator variable, we found that the relationship between right posterior hippocampal volume and cognitive map accuracy was moderated by participants’ self-report sense of direction and route integration ability. Such moderation was not found in the right anterior hippocampus. Interestingly, this result was more clear with SOD than Silcton grouping. One possible explanation is SOD is a continuous variable and was free of arbitrary grouping. The Silcton grouping, on the other hand, was very well motivated for its original intent, but for our analysis it discretizes ability and was based on the pointing task performance which was highly correlated with the Model building task performance (given that both tasks probed the spatial representations of the same environment). Such collinearity between the Silcton grouping and the model task performance might contribute to the absence of a moderation effect on the Model building total measure.

Weisberg et al.’s (2019) data prompted us to evaluate how researchers have conceived the relationship between hippocampal volume and spatial learning performance. The theoretical foundations of our re-analyses come from two studies (He et al., 2019; Weisberg and Newcome, 2015). Both of these studies suggest that “good” and “bad” navigators at test reflect a confluence of factors including flexibility with cues and effective strategy use alongside memory capacities that may be supported by the hippocampus. Note that the flexibility we refer to here may be limited to spatial learning contexts, as “bad” navigators could still in principle have great flexibility in solving non-spatial tasks. It remains an open question why “good” navigators appear to have greater flexibility (Wolbers and Hegarty, 2010) as research on this topic is sparse. Here, we put forth an idea in this paper to be pursued with continued research that perhaps either group of navigators could encompass individuals with variable memory resources, and their observed spatial learning performance reflects whether they utilize resources (including external - environmental cues, and internal – cognitive/neural resources) differently and in the optimal manner for the navigation tests they are ultimately given. This difference of resources utilization could lead to very different learning outcomes for these two groups of people: from our prior study (He et al., 2019) we hypothesized that only good navigators would take effective advantage of the resources to facilitate cognitive map formation. The results of this reanalysis suggest the distinction may be more extreme in tasks like Silcton, with poor CMF performance possibly emerging from using greater hippocampal mnemonic resources to learn details of the task in a manner that proves maladaptive for the subsequent testing.

If our idea holds up in subsequent studies, one promise of this framing is the fact that spatial strategies can be changed within individuals (Boone, Maghen, & Hegarty, 2019) – and therefore it may be that real-world navigation performance could be improved for many through interventions that target, for example, spatial strategy, attentional allocation, and possibly appraisal of one’s abilities. It also makes the prediction that people can better allocate their resources when they know what aspects of the environment they will be tested on, and then reveal the limits of their spatial abilities (dissociated from strategy preference – e.g., Weisberg and Newcombe 2015).

Our reanalysis suggests that researchers should model participant attributes such as SOD to better capture the correlation between hippocampal volume and spatial learning performance. Why then have previous studies in healthy young adults still shown significant correlations even though they did not subdivide participants (e.g., Brown et al., 2014; Chrastil et al., 2017; Sherrill et al., 2018)? Based on the above discussions, we conjecture that the mixed findings could be due to whether 1) tasks enable different strategies or informative cues to be adopted, and 2) participants had an opportunity to adjust their spatial learning strategy based on the observation of test demands. Indeed, in earlier studies wherein learning and testing were repeated in different environments (Brown et al., 2014; Chrastil et al., 2017; Sherrill et al., 2018), participants had opportunities to adjust their learning strategy in the next learning-testing cycle based on the test they received in the previous one. In some designs like Weisberg et al.’s (2019) experiment, however, participants did not know how their spatial knowledge was to be tested and had no opportunities to adjust their default encoding strategy to reflect their abilities with the testing demands. In other words, in many previous studies participants could adopt the strategy and reference frames they saw fit based on observed task demands, but Weisberg et al.’s (2019) study importantly targeted the CMF learning outcomes from a participant’s default strategy and encoding tendencies. If our speculation is true, another hypothesis to be explored is whether hippocampal volume may not only provide greater relational memory resources, but through this engender greater flexibility in how people can adapt their learning strategy to navigation demands (indeed, this was an explicit challenge for learning in Brown et al., 2014). Additionally, some tasks may be solvable using different strategies or cues with similar efficacy (e.g., maze-routes; Brown et al., 2014) such that greater mnemonic resources (not, e.g., spatial reference frame) are the primary limiting factor for observed “ability”. We hope that future work tests these ideas.

Except for the significant moderation of SOD on the relationship between right posterior hippocampal volume and model building task performance, we did not observe such moderation effects on other relationships, including anterior hippocampal volume – model building task performance, or right hippocampal volume - pointing task performance. These null results support Weisberg et al.’s (2019) findings that the volume of the medial temporal structures (hippocampus, parahippocampus, entorhinal cortex) did not have an overall significant correlation with spatial learning performance (pointing and model building tasks). It is natural to ask why the relationship between hippocampal volume and spatial learning performance, moderated by SOD, localized to the right posterior hippocampus and the model building task. In rodents, it has been shown that the dorsal hippocampus, which corresponds to the posterior hippocampus in humans, is more strongly associated with spatial cognitive functions and information processing than the ventral hippocampus, which corresponds to the anterior hippocampus in humans, bolstered by connectivity and genetic differences (Fanselow and Dong, 2010). It has been suggested that the human hippocampus could have a similar division of labor (see evidence from Libby, Ekstrom, Ragland, & Ranganath, 2012). In terms of the spatial task differences, it is interesting to note that performance correlations between the pointing and the model building tasks are between 0.46 to 0.58 in Weisberg et al. (2019), which – although significant - are considered to be between low to moderate (Hinkle et al., 2003). These correlation magnitudes on one hand clearly suggest that the pointing task and model building task tap into overlapping constructs underlying spatial memory, but on the other hand they also suggest that distinct aspects of spatial memory are sensitive to each of the tasks. We focused on the model building task because we hypothesized it may relate to the concept of a cognitive map more closely, but importantly this does not mean that pointing tasks are less important assays of spatial cognition. Instead, as discussed in the Introduction, the different reference frames (allocentric vs. egocentric) encouraged by these tasks might be a key difference. It is important for future research to investigate whether the nature of reference frames could moderate the relationship between hippocampal volume and spatial learning.

Although our re-analysis is motivated by previous studies, the number of studies targeting individual differences within navigator sub-groups is relatively scarce and the connections to the present study are indirect. Compared to the pre-registered nature of Weisberg et al. (2019), our re-analysis is inherently exploratory and the weight of evidence must first be given to the outcomes of the pre-registered tests on these data. Another limitation of the present work includes the smaller sample size when participants are divided into groups, although this problem is mitigated by the SOD moderation analysis which capitalizes on the full continuous data. We found Weisberg et al.’s (2019) conclusions extremely thought provoking, and it is our hope that researchers utilize ideas such as ours to develop studies that look into this topic further before considering it ‘cut and dry’. The current re-analysis does not attempt to reduce the importance of Weisberg et al. (2019) – indeed, we argue additional evidence for their speculation that the relationship between hippocampal volume and cognition may be indirect. Rather, we hope that our analyses provide concrete future research plans which include a) treating individual’s spatial ability as a moderator, and b) designing studies with two navigation tasks – one which allows spatial strategy changes, and one which does not, enabling us to explore the idea of strategy+reference frame flexibility and its possible relationship to morphology. As noted by the late Howard Eichenbaum, “hippocampal networks [can] map multiple navigational strategies” (Eichenbaum, 2017). We suggest this relational memory perspective holds even in instances where not all individuals fully (or optimally) utilize the available resources. Lastly, the different patterns of correlations among individuals shown here further support the long-held view of a complex relationship between brain volume and cognitive outcomes (Rushton & Ankney, 1996).