Some of my previous posts have considered how Auckland compares to the Netherlands in terms of general consideration for pedestrians. Based on the results of this discussion it seems fair to suggest that the largest divergence in engineering practices between Auckland and the Netherlands occurs in the following areas:
- Left-turning vehicles in Auckland are often provided with slip lanes that allow them to maintain high speeds even as they turn.
- Pedestrian crossings in Auckland are commonly removed in order to expedite vehicle flows (with little to no regard for how pedestrians can subsequently cross the road).
- Vehicle lanes in Auckland are very wide, which in turn increases vehicle speeds and the distances pedestrians must walk to cross intersections.
Taken together this means that 1) vehicles travel relatively fast in Auckland and 2) walking in Auckland is relatively difficult and unpleasant.
In response to my wailing some people suggested that Auckland’s preferential treatment of vehicles was warranted because relatively few people in Auckland actually walk, at least compared to places like the Netherlands. And if this were true then this may indeed warrant different priorities. After all, it would not be prudent to invest large amounts of resources (either directly in the form of infrastructure or indirectly in the form of additional delays to vehicles) in order to meet the needs of relatively few pedestrians.
But is it true that Auckland has low rates of walking compared to the Netherlands? To investigate this issue I collated some data on journey to work mode share for several European cities, which I then compared to the walking mode share in Auckland City, as illustrated below.
These journey-to-work mode share statistics reveal something quite special: Auckland’s walk mode share is actually twice that found in Amsterdam. This statistic on its own surely puts lie to the suggestion that Auckland’s inferior treatment of pedestrians is somehow related to less demand than is found in the Netherlands. In fact the reality is quite the opposite – within the former Auckland City area we actually support higher rates of walking than several, much larger, European cities, such as Berlin and Hamburg amongst others.
As with most analyses, however, this result just raises a further question: If Auckland does indeed treat pedestrians so poorly (as I have suggested) then why are rates of walking not lower than all these other cities? It’s a good question, but one that may also have an answer. My hypothesis in response to this question is this: Notwithstanding Auckland’s poor pedestrian infrastructure, we are still able to sustain relatively high levels of walking because of our reasonably favourable climate.
In order to test my hypothesis I collated some additional climate statistics, namely annual precipitation, annual sunshine hours, and annual mean daily temperature, for all the cities included in the table above. These relationships are illustrated in the following three graphs.
These results suggest a relatively strong positive correlation exists between walk mode share and mean annual temperature (R2 = 11%) and annual sunshine hours (R2 = 32%). There also seems to be a strong negative correlation between walk mode share and annual precipitation (R2 = 24%). If you look closely you can see that in all three charts I have highlighted the point for Auckland in red. This in turn shows that Auckland is below the trend (compared to other cities) in terms of the level of walking we support based on temperature and sunshine hours, but above the trend based on our levels of precipitation.
But of these three factors which is the most important? To answer this question we need to move from the single variable relationships shown above and instead analyse multiple variables simultaneously. My multi-variate regression results are summarised in the following table (NB1: who ever said economists have no sense of visual aesthetics?!? NB2: Econometric wizardy such as this is something that all transport planners should try to pick up – it’s incredibly useful).
For those who are not accustomed to the beauty that is regression analysis let me clarify these results as follows:
- The overall “goodness of fit” of the model is indicated by the “R-squared” variable, which in this case is 36%. This suggests that our regressors (namely “rain” and “sun”) are able to explain 36% of the variation that is observed in walk mode share. Stated differently, 64% is explained by other factors. This is actually not to bad because there’s a lot of factors aside from climate that can influence walk mode share (which is discussed in more detail below).
- The left hand column lists our variables. Walk is the dependent variable, i.e. it is the variable we are trying to explain in terms of the regressors, namely “rain” and “sun”. The “cons” variable is simply the constant of regression and is meaningless (it just describes the hypothetical walk mode share one would expect in the event that you lived in a city that had both zero rainfall and zero sunshine).
- The “Coef.” column signifies the direction of the relationship between our dependent variable (“walk”) and the regressors. So we see that walk is negative related to “rain” and positively related to “sun”. More specifically, a 1mm increase in annual precipitation reduces walk by -0.00402%, whereas a 1 hour increase in annual sunshine increases walk by 0.00288%.
- The P>|t| column essentially describes the probability that our regressors are not statistically significant. We see that there is a 17% and 3.8% chance that our rain and sun regressors are not statistically significant. Thus the balance of probabilities suggests there is a statistically significant correlation between walk and our chosen regressors.
Note that in this regression I have omitted temperature because it is (as you would expect) strongly correlated with sunshine hours – as such including both temperature and sunshine hours as regressors would really just dilute their individual effects to the point where neither were statistically significant.
The final step is to use these regression results to “predict” the walk mode share for Auckland. This is calculated simply as follows:
Walk mode share (Auckland) = -0.0000402*Rain + 0.000028*Sun+0.0765143 = -0.0000402*1240mm+0.000028*2007hrs+0.0765143 = 8.45%
So how does our predicted mode share compare to actual mode share? Well, the latter is 7.84%, which is not too far off what we have predicted above. Nonetheless, when considering Auckland’s climate we would actually expect an even higher walk mode share than what we actually have, holding other factors constant. Auckland is, based on this analysis, actually under-performing slightly in terms of our walk mode share, at least given our relative climatic advantages.
But what other variables (not included in our model) might explain Auckland’s slightly lacklustre walk mode share? Some obvious ones include:
- Poor pedestrian infrastructure, as mentioned at the start of this post;
- Auckland’s relatively low average density (2,900 peeps per sqkm versus an average of 3,900 peeps per sqkm for the cities in our sample);
- Geography, namely our relatively hilly central city area.
But I’m sure people out there have other their ideas of their own, which I’d be interested to hear about. Before we end this post let’s just recap what this all means:
First, Auckland seems to have relatively high levels of walking when compared to similar sized cities in Europe, which in turn suggests that our poor treatment of pedestrians is unlikely to be explained by underlying differences in demand. Instead, it seems more likely that our relatively poor pedestrian infrastructure reflects underlying differences in our traffic engineering practices – the needs of pedestrians simply are not given as much weight in New Zealand as they are overseas. And I think that’s a real shame.
Second, and in a more positive vein, Auckland seems to have some natural climatic advantages that support high levels of walking. That suggests to me that if we are able to get our traffic engineering practices on par with those found in these European cities (i.e. avoid slip lanes, install pedestrian crossings on most approaches to intersections, narrow the width of vehicle lanes, provide raised pedestrian tables at intersections, and improve general street amenity) then we can look forward to even higher walk mode share in the future.
There you go – we did end up finishing on a positive note :).
Processing...
A very interesting analysis. The key question here is – “are European city dwellers simply walking less due to better public transport facilities and a greater cycling share” as well. I think, given that the modal split is generally more veered away from cars in those cities citied as having lower walking shares than Auckland, that there is simply a greater preference towards using modes other than walking. Given that Berlin, for example, is a vast, sprawling city, the percentage of people who can simply walk to their place of work is innumerably lower than Auckland despite geographical dispersion evident in Auckland too. Having experienced a few of these European cities cited first hand, and given that I live in Stockholm now, I can quite safely venture that walking shares are only such in cities where home, work and play are closely intertwined.
Stockholm, to use a case study, started the ABC concept for its suburbs in the 1950’s (Arbete [work]. Bostad [apartment], Centrum [Centre]) meaning that everything was meant to be self-contained in the suburb. The suburb of Sundbyberg, for example, requires by law to this day that there is one job available for every household in the suburb. This practice alone will of course increase walking share as everything will be much closer to your dwelling. If one compares that to other rather more centralised cities, or cities that rely upon isolated business parks one can see a precipitous drop in walking and a commensurate rise in other modes the more one isolates areas into “single use” areas. So whilst analysing pedestrian friendly infrastructure is useful, one should also analyse the total urban fabric of the cities in question in an attempt to explain why cities with generally poor climates (like Stockholm) can still exhibit high walking shares and the lowest car use of all cities cited above.
Hejsan “Black Metal” välkommen in. Du får gärna snacka lite svenska om det passar på.
Very good point – and reset assured I’ll be exploring some of the other interesting lessons from this data in future posts, i.e. vehicle mode share and other urban form attributes.
Tack så jättemycket. I will keep to English seeing as that is my native language (native Brit who lived in Auckland for 7 years before moving back to Europe). I’ve followed the blog for a long time and know Josh, but I decided to start posting here given there are a number of very stimulating discussions under way and this is the best source for Auckland transport discussion I’ve found on the web.
Keep up the posting and I look forward to further posts in this vein.
Interesting analysis Stu, and while I don’t want to seem to be knocking it and think it’s a useful start, I think it would benefit from comparing the Auckland region/metropolitan area, rather than Auckland City area – I assume this data wasn’t available? To me this analysis isn’t comparing oranges with oranges but maybe mandarins and oranges, similar but still quite different? I would assume, though have no figures to back it up, that the 8% figure would decrease once other areas further away from the CBD were added in.
Also on another note it’s important to remember how important the pedestrian environment is to PT – we can create the best PT system in the world but it won’t perform as well as it could unless the walk to the stop/station is pedestrian friendly.
Hi Al C – good comments, my response:
1. I did not jump to the regional level because the Auckland Region includes large tracts of rural land that is not comparable with European definitions of cities, such as Amsterdam, which are much more associated with just the urban area. In terms of apples and oranges that’s really part of the game: No administrative definition of cities is ever the same, but I would not jump so readily to the regional level and assume it’s comparable, because the reality is that the definition of cities in Europe will generally not incorporate areas like Franklin, Rodney, and even the Waitakere Ranges. So maybe I drew the urban boundaries too narrow, but the regional boundary is on the other hand too wide I’d suggest.
2. I did, however, just do a sensitivity test which included the old North Shore, Manukau, and Waitakere City Councils. Walk mode share did indeed drop to 4.9% – but remains above that found in Amsterdam so still not too bad really (although much worse than one would expect considering our favourable climate). This suggests that even across the wider urban area we have levels of walking that should sustain better infrastructure. And I’d also ask you this question: Given Auckland City was, for many years, an administrative unit in its own right, why is the pedestrian infrastructure not better there? Stated differently, Auckland City obviously justified better pedestrian infrastructure on it’s own. I could do a more fine-grained analysis of walking by CAU but at 1.30am last night I thought it was time for bed :).
3. Completely agree with you on the link between walking and PT. The former is likely to complement the latter.
Walking infrastructure is poor in Auckland City because for years we have had the cit rats running the show. Don’t forget, it was Banks’ first council that happily accepted the Birch reported which stated that walking and cycling should be discouraged because they didn’t pay petrol tax.
You make a very good point here and I think that Singapore is the perfect example of having a very good public transport system, but having very, very poor pedestrian provision. Singapore is very pedestrian unfriendly in the suburbs with very limited crossings and often obstructed pathways with trees, large drains next to the path, pathways ending suddenly and transferring to the other side of the road not to mention the large fines for jay walking. If pedestrian facilities in the city were improved, then I am fairly certain that one would see a commensurate increase in public transport patronage.
Yes, from what I’ve heard and read Singapore seems to treat pedestrians extremely poorly. On one Singaporean blog I even read a comment that implored Singapore to improve its public transport system because otherwise people would have to walk and that would lower productivity (presumably because of the additional time it would take)?
I’m fairly sure that my health and productivity is better because I don’t use public transport to get to work but instead walk/cycle.
Singapore would also not fit your relationship between sunshine hours and conduciveness to walking- many places in the world have a bit much of this good thing.
It might be more of a U shaped curve. Once the sunshine hours and heat get too extreme (think Saudi Arabia, south of Spain) people barely move.
Hi Al C, good comments; my responses:
1. The reason I did not jump to the regional level is that Auckland Region includes large tracts of rural land that is not comparable with European definitions of say Amsterdam. In terms of apples and oranges that’s really part of the game: No administrative definition of cities is ever the same, but for the reasons stated in the last point I would not jump so readily to the regional level and assume it’s comparable, because the reality is that the definition of cities in Europe will generally not include areas like Franklin, Rodney, and even the Waitakere Ranges. So maybe I drew the boundaries slightly too narrow, but the regional is too wide I’d suggest.
2. I did, however, just do a sensitivity test which included the old North Shore, Manukau, and Waitakere City Councils. Walk mode share did indeed drop to 4.9% but remained above that found in Amsterdam so still not too bad really (although much worse than one would expect considering our climate). This suggests that even across the wider urban area we have levels of walking that should sustain better infrastructure. And I’d also ask you this question: Given Auckland City was, for many years, an administrative unit in its own right, why is the pedestrian infrastructure not better there, even if it could be worse elsewhere? Stated differently, Auckland City obviously justified better pedestrian infrastructure which even in pedestrian dominant areas it fails to provide. I could do a more fine-grained analysis of walk rates by CAU but did not have the time …
3. Completely agree with you on the link between walking and PT. The former is likely to complement the latter.
Great stuff Stu, ah but stats, eh? One issue is that every PT user is, to varying degrees, a walker. Of course not for their whole journey, but still requiring good pedestrian amenity. Interestingly cycling and driving share a higher likelihood of being more point to point. So while I take your climate point and its conclusion- we could up that mode share enormously, this data is only about whole journeys.
There is no doubt that we get what we invest in; ever been tempted to walk along Newton Rd or other motorway blasted routes? No me neither.
Ed Glaeser is very big on climate, claiming that temperature is the single biggest determinant in city success over the last 60 years…. since the spread of air conditioning in other words. Certainly works in NZ too; Dunedin used to be our biggest city. This may be changing as the high water/electricity/fuel costs of the spread out hot cities of the southern states start to loose their appeal.
Yes, as energy costs rise relative to historical levels you would expect cities in temperate climates (not too hot, not too cold) to enjoy a relative advantage over other places. As you would expect cities that are less vehicle dependent to enjoy a comparative economic advantage. Shame our Government does not realise how much their transport policies are undermining Auckland’s energy resilience …
There seems to be some inverse relationship between walking and cycling mode share which is interesting. In Amsterdam the cycling facilities are so good that people cycle for even short journeys that in other cities people would walk for. Another point in why walk share is higher in Auckland is that public transport is bad, so this pushes more to walk for short journeys. However as Patrick mentions above improving public transport will increase walking km’s as people walk to and from stations.
Yes well-spotted; the relationship suggests that for every 1% increase in cycling mode share there is a 0.36% decrease in walking mode share. Put another way, approximately 1/3 of people who take up cycling would have otherwise walked. That’s actually a fairly standard kind of diversion rate for modes that are relatively substitutable.
A similar (although much stronger) relationship holds between vehicle mode share and public transport mode share: Every 1% increase in PT is associated with a 0.8% reduction in vehicle mode share for these cities. That suggests that 4/5 people who use PT would have otherwise driven.
And now you know what the next post is about :).
Please be cautious about doing correlations with proportions, if that is the plan. Because these proportions sum to 1 they are by definition non-independent. In fact, they are ‘compositional’ data.
Consider that if you aggregated to just ‘Car’ and ‘Other’ then you would find a perfect negative correlation between the two since Car + Other = 100% for all cities. The problem is barely reduced by using four categories. And especially be cautious about interpreting correlations between proportions as ‘causal’ in the way that this post suggests – these data simply can’t support claims like 4/5 people who use PT would have otherwise driven if that claim is based on a -ve correlation of around 0.8 between the two.
In Auckland about 80/(100-11) = 90% of people who use PT could otherwise be expected to drive (that’s what the overall proportions suggest), but in Barcelona 39/(100-46) = 72% of people who use PT could otherwise be expected to drive. Those numbers aren’t miles from your 4/5, but that’s just a coincidence. And in Auckland 8/(100-2) = 8% of people who cycle would otherwise walk, and 2/(100-8) = 2% of people who walk would otherwise cycle. There’s also the question of how relevant the average relationships between modes across a bunch of cities are to the situation in a particular city?
Anyhoo… using these numbers to derive correlations between proportions that sum to 100% can lead to all sorts of erroneous conclusions! There is a set of approaches that can cope with compositional data like these https://en.wikipedia.org/wiki/Compositional_data – Aitchinson is the man… but it’s not as much fun as regression models.
Yes good points – but these kind of mode share modelling issues are dealt with in the literature. The simplest approach is to apply a logistics transformation of the dependent variable (see Small and Verheof, http://www.amazon.com/Economics-Urban-Transportation-Kenneth-Small/dp/0415285151).
I could be wrong, but I don’t think that the logit transformation gets you off the hook if the idea is to use one mode-share as a predictive variable for any of the others. It’s fine to apply the logit transformation to a dependent variable (that’s what logistic regression is, isn’t it?), but you should steer clear of using the shares to predict the other shares!
Two points Stu. One, look at the cycling rates in Amsterdam compared to Auckland. Why would you walk when it is so easy and safe to cycle? Secondly, it would be interesting to compare pedestrian accident rates between Amsterdam and Auckland.
Yes – see my comment in response to Luke above. But I think the main point still holds (correct me if I’m wrong) that in Amsterdam they provide for both pedestrians and cyclists, even though the former are not a particularly large number of people. So that suggests that traffic engineering practices have made conscious decisions to provide for pedestrians differently from how we provide for them. If that makes sense?
Complete sense Stu. I see where you are headed with this and I like it :-).
As an aside, I read a document the other day and the Dutch have quite a rigid criteria on types of roads and the subsequent treatment of cyclists / pedestrians based on the type of road. As an example, I do not think you would find an uncontrolled crossing on a 4 lane / 50 km/h road in the Netherlands, whereas we have them all over the place. Appropriately enough the Netherlands has the lowest accident rates for pedestrians in the EU.
The other thing I have noticed is that the Dutch have a very uniform design for pedestrian crossings for the entire country as compared to the completely random designs in New Zealand. I was told about a crossing in the South Island the other day that just had different coloured pavers as the markings for the crossing. Very ambiguous and dangerous.
Yes, the Dutch just seem to have better traffic engineering practices.
So those bike bumper stickers that say “one less car” should really read “one less pedestrian”?
Almost – 0.36% less pedestrian, which you would round down to zero if you were specifying to one significant figure 😉
Strictly anecdotal: I went from driving to cycling [when commuting]. Tried the bus first, and still use it occasionally but too slow and indirect [020, the Link is better], train for longer journeys occasionally as well as driving [the only fully supported mode in AK]. Point to point cycling is clearly the fastest for my commutes; Grey Lynn to city or University. Also cheapest, most tiring, most exciting, and almost certainly most dangerous. It is hilly which is of course a good workout [I am fitter now], as well fun going down, but not for everyone and it does make you think about what you’re carrying [no good with a camera case], and worry a little about personal hygiene- I prefer to wear real clothes, this is transport after all, not training for a race.
Car and petrol use is down significantly, like most Aucklanders of my generation I used to only ever drive everywhere; when on my bike I really am one car off the road. Our household’s next change will be to go down to one car. Which incidentally is one reason inner city property can command higher prices.
Along the way there is lots of opportunity to consider street life + design, architecture, and of course driving habits…. One curious thing about being on a bike in traffic [I’m always in traffic] is that you have a generally elevated sightline to people in cars and therefore have a great view in: FYI, every single ride I will see someone driving with a phone to their ear.
Individual anecdotes welcome :). I think that if there was one mode where Auckland clearly has opportunities to lift mode share it is cycling, closely followed by public transport. Mainly because the “natural” market share for these modes must be higher than it is currently, based on what is achieved in other cities.
Jeepers creepers even Helsinki achieves a cycle mode share of 6%, which is three times Auckland’s, despite being one of the most frigid cities on Earth. And cycling in snow is not much fun I can tell you that (says the man who flipped off his bicycle twice on ice when in Amsterdam). Auckland could get so many more people cycling …
This is fascinating work, thank you for doing this. A couple of comments, with the caveat that my knowledge of stats is rather limited. Someone like Xavier knows far more than I do 🙂
I certainly wouldn’t consider a r of .11 as moderate. It’s a pretty weak correlation. You can see quite obviously that many cities with lower temperatures have high walk share, despite some relationship. Same goes for .3 I think you’re overstating the relationship a little – though it’s worth describing.
Your sample of mostly Spanish, Dutch and German cities is arbitrary, and there’s nothing wrong with that (the set of cities is essentially infinite for our purposes and a selection should be made. But that Auckland is an outlier in precipitation (eyeballing the data I wouldn’t be surprised if Auckland was more than 2 standard deviations from the mean) means that your data isn’t really comparable and the question you’re asking – how much does precipitation affect walk-share in Auckland – won’t be answered with this data. To fix this I certainly suggest that you include a number of cities from developed countries which have similar rainfall patterns – perhaps Singapore, Taipei, Sydney, Wellington etc. Similarly, while mean annual temperature tells us something, it doesn’t adequately describe how Auckland relates to these cities, many of which have very disimilar temperature patterns (cold European winters and hot European summers). It also looks like you’ve essentially got two groups in your sunshine data, which given the geographical distribution of cities (the group on the right being from Spain) makes absolute sense. That weakens your predictive power.
Adding a number of temperate cities across a more diverse range of geographical and climate locations would help resolve your problems. So, use more data points with less homogeneity 🙂
George, I think your intuitive knowledge of statistics seems very good! My responses as follows:
1. An R2 squared of 0.11 is actually pretty good for a single climatic variable describing a complex socio-economic phenomenon like walk mode share. It’s a different story when you are doing controlled experiments, such as what tends to happen in the biological and behavioural sciences, because in that case you have much more control over the environment. What’s more relevant is the statistical significance of the relationship, which considers the correlation and the sample size. E.g. an even lower R-squared can often be statistically significant if you have a large number of data points.
2. Building on the last point, what the results of the multiple regression analysis suggests is that two climatic variables on their own explain 36% of the variation in walk mode share. I think that’s rather high when you think about other potential factors that will also have an impact, such as demographics, vehicle ownership, and land use patterns. The general message is that low R-squareds are acceptable when dealing with messy socio-economic data, or when you have a large sample (not they I do), and/or you’re using a simple model that omits many other variables.
3. My sample is definitely limited and potentially biased. I have relied on data available here: http://www.urbanaudit.org/DataAccessed.aspx, but added Auckland. I would not be opposed to expanding the sample, but again my time last night was limited so I took the easy option. Please send through more data if you have it!
4. In terms of controlling for country differences, the simplest solution would be introduce a dummy variable for countries and/or groups of countries to control for common cultural effects. In this case I think that would best way of resolving the issue of data subsets and controlling for heterogeneity linked to national policies/preferences.
Regards,
Stuart.
Correct me if I’m wrong, but is your multiple regression model done in Stata?
Yes it is.
Heh – I thought I was the only person who used Stata.
The NZ regional stats for mode share definitely confirm the rainfall hypothesis. Most windward regions have less than 4% mode share for walking/cycling whereas the leeward regions are mostly above 10%. Population density appears to only be a significant factor for pt mode share. Since the stats are for travel to work one can speculate that small towns are more sustainable than big cities because a greater proportion of workers live within walk/cycle distance of their workplaces.
http://www.transport.govt.nz/ourwork/TMIF/Pages/TP006.aspx
yes well-spotted Kevin – this is something I have looked at previously and came to the same conclusion. In this post I was more interested in international comparisons because I wanted to partly debunk the notion that nobody walks in Auckland compared to cities overseas. But investigating whether the same climatic relationships hold within New Zealand would be an interesting “out of sample” test of whether the relationships were relatively universal (and if they are then they have that much more weight).
Just one thing: I would not necessarily conclude that small towns are more sustainable than big cities, because personal transport energy demands are only one part of the wider equation, i.e. there may be other economies of scale in ecological impacts that are associated with larger cities. Although maybe your comments were intended to apply only to sustainability in a transport context?
Stu, it seems like it would be good to put together the mode-share arguments you’re making with the argument Peter makes that the marginal cost of adding capacity for rush-hour cars is very high ( http://greaterakl.wpengine.com/2012/08/04/the-future-of-fuel-taxes/ ), and end up with a compelling argument that quite substantial investment should be made to in pushing walking/cycling as long as it results in even small peak hour mode swap?
Interesting suggestion. My argument for investment in walking facilities is really based on notions of ethical values – pedestrians are people too and they deserve considerably more attention than they’re currently getting. But yes, you definitely could make the argument (and I have in the past) that even relatively small increases in walking/cycling mode share would deliver quite high congestion benefits at the margin … I’ll think about this …
Mostly this is a nice analysis, but I’m pretty sure the likelihood of overfitting is exceptionally high in the multiple regression analysis. There simply aren’t enough data points to be able to tease out more complicated relationships.
By “over-fitting” I assume you are referring to there being relatively few degrees of freedom (n = 22 – 3 = 19)>
Maybe – the small sample size tends to affect the statistical robustness of the results, rather than their socio-economic significance. So I’d push back on your suggestion that the likelihood is “exceptionally high”. I’d expect that more data points would reduce the P-value and sharpen the estimates of the coefficients, but it should not change the general direction of the relationships between climatic conditions and walk mode share. That is, I would always expect that more precipitation is likely to mean less cycling, and vice versa for sunshine hours, holding other factors constant. You can corroborate this finding by looking at seasonal cycling statistics in Auckland – where rates drop off in winter months. So I think you’re right to suggest that more data points would be good, but wrong to suggest that the few degrees of freedom would fundamentally alter the results.
I love statistics.
I dunno – the somewhat bodgy (technical term) rule of thumb I was told is that for multiple regression you need at least 10 observations per independent variable minimum, so on that basis the analysis just squeaks in – it’s barely ‘ multiple’ regression after all! Where I’d differ from Stu’s interpretation is that the t test scores for individual coefficients suggest that precpipitation has no clear effect once sunshine hours are taken into account. Perhaps that is another correlated independent variable – given the small sample it wouldn’t be at all surprising if (Auckland apart!) the driest places here were also the warmest ones. [George has also made many good and related points above]
On the meta-message… Glaeser’s emphasis on climate in ‘Triumph’ is pretty interesting although reading it in mostly climatically dodgy and super-successful Tokyo I found it less than persuasive, and smacking a little too much of environmentally determinist explanations for why modern science originated in Northern Europe and not in sub-Saharan Africa. I think economists are prone to waving regression model around as a ‘sophisticated’ methodology but then often fail to back up the claims with plausible explanations that stand much scrutiny (cf Freakonomics, which is entertaining, but nowhere near as smart as it thinks it is. The Spirit Level is better, but then they take more time to construct believable explanations – probably why their critics focus on the regression models and not the message…).
Ultimately, as one must always tell students: correlation is not causation. That London had first-mover and then imperial capital advantages based on the piratical trading activities of its navy and slave trade followed by positive feedback effects and lock-in (i.e. path dependence) is a rather more convincing explanation of its longstanding success than, “weather mostly harmless”! Beijing, or Calcutta, or Cairo, or Sao Paulo or wherever (insert your alternative history here) could just as easily have been that city, and then maybe he’d be saying that hotter or wetter cities were more successful.
Ahem… to get back to the OP, there is little doubt that climate and topography affect people’s propensity to walk, but I’m not convinced that regression is a sufficiently subtle tool for drawing that conclusion from these limited data. Besides which, for me, the glaring difference between these cities is in the modeshare of cars – are we to conclude then that warm, wet climates cause more people to drive? 🙂
I don’t disagree with the intent here – Auckland could be one helluva lot more walkable, and there really is no good reason not to make it so!
All good points, but I would just like to point out that econometrics, or regression analysis, is never a substitute for theory or even intuition – and the latter two should come first before you even delve into the data, as I have presented them in my post.
What I hypothesised was that:
1. If you live in a rainy climate you are less likely to walk to work; and
2. If you live in a sunny climate you are more likely to walk to work.
To me that all seems very reasonable and subsequent empirical analysis suggests (this is a very key word) that the data does not disprove these hypotheses. Of course, the relationships are not statistically strong – because of the small sample size – but nor can they be ignored (just as stylised facts about Melbourne being a great city and having laneways cannot be ignored, even if it should be interpreted with caution).
The trick now would be to go beyond my analysis of these cities by either: 1) Looking at more cities 2) Looking at more cities over time (i.e. does walk mode share go up in Auckland in dry/sunny years?) and 3) Looking at whether the relationships hold outside of this sample (i.e. do the same climate variables explain walk mode share in cities/towns within New Zealand?). Basically what I’m trying to say is that regression analyses never “prove” anything – all they do is define the level of confidence you should have in your hypotheses, or intuitions, by disproving whether you were wrong.
With regard to correlation versus causation: It’s important to note that while correlation does not prove causation, the absence – or lack – of correlation usually disproves causation. That is, if you hypothesize a relationship but find no correlation between key variables, then you’re probably either a) measuring things wrong or b) not onto something. So at least here I’ve found that I’m not barking up the wrong tree; I’ve just narrowed it down to a choice of 5,345 different trees ;).
Sorry, I’m an SPSS man, and was slightly guilty of skim reading your post. SPSS lists the intercept/constant first in its output. If you had had two significant predictors variables, with an n of 22, the second predictor is highly likely to be spurious. So I withdraw and retract my initial criticism, it’s not valid here. The risk is that the second relationship may be entirely driven by a signal datapoint, but that that is not obvious unless you start graphing residuals.
Walking speed is a good indicator of the adequacy of the infrastructure supporting walking.
“Field observations of factors influencing walking speeds” K.K.Finnis and D.Walton.
Not sure about Kevin’s regional hypothesis. By way of anecdotal rebuttal I’ve just spent 30 hours in Hamilton which, from my not exactly limited experience (I’m 55 and have spent most of my adult life living either in Europe – Belgium, Denmark, the UK – or Australia), is probably the most pedestrian unfriendly place in the world. It shouldn’t be: it’s largely flat, has a relatively small CBD and a large student population. Unfortunately, over the past thirty years, this CBD and the city’s inner suburbs have been progressively demolished in order to build car yards, car parks and retail superstores. The roads are wide and there’s a history of constructing failed by-passes, etc, to deflect cars from the main street; there are numerous roundabouts (with expanded necks from the approach roads); but there is not one zebra crossing. Timing on pedestrian beg buttons is uncommonly long, where such things exist and there are numerous intersections that only allow pedestrians to cross in two out of four possible routes. The consequence, of course, is that the city is devoid of any pedestrian component (except for outside the main Work and Income office on Victoria Street) and consequently any sense of urban buzz. Hamilton has effectively driven itself to death.
Oops, sorry, Kevyn, not Kevin.
Christchurch plan looks a lot like Hamilton
Have you TRIED walking in daytime temperatures in Singapore? It is not for the faint hearted! 🙂
Correct me if I’m wrong but these stats are based on the % of people who walk to work. Does it take into account how FAR they have to walk? European cities are far more densely populated than Auckland so their walk to work is down streets well protected by high buildings, not a 1km treck from Parnell alongside the container terminal in horizontal rain
Yes, Stu, my comment was specific to personal transport, and especially to towns well away from cities. Spreading all of a city’s industrial functions between hundreds of small towns would be a sustainability nightmare. With high-speed bradband it might be possible to spread many administrative functions across hundreds of small towns sustainably, although some might object to only being able to do virtual lunch meetings at virtual restaurants 🙂
Kevyn your last point is not as light hearted as the smilely-face implies. Face to face contact still matters regardless of your bandwidth. Ed Glaeser is very clear about this, and its not just about work either, entertainment is just as important for a place to be competitive; just try recruiting smart ambitious types to work in a sleepy backwater…..
ok. So I’m very late here but I have inadequate internet access right now. anyway, just the brief comment I wanted to make about this is that at one point in my life for a psychology assignment I ended up reading a great deal of article about how weather in particular cities or countries affects peoples’ behaviour. There has been a lot of research done on this topic, particularly in regards to how often people have sex and how likely they are to commit violent crimes.
So, basically, what I got from reading these articles, particularly the ones about sex, was that a) some social psychologists just have WAY too much time on their hands and b) relationships between weather patterns and behaviour are often more complex than you would expect. In fact, in some cases they are so complex that you get quite sceptical about whether there actually is any relationship, or if they are just going on “fishing trips” for significant relationships.
Of course, that may not be true of walking because it is obviously a behaviour that is more affected by weather than most of the others that have been studied a lot. But I do think it might be interesting to look at some of those studies because they really do get quite complex in terms of looking at things like how extreme temperatures affect behaviour (they are often more powerful influences on behaviour than averages), and also things like cloud cover etc.
One thing I wonder about Auckland is whether the unpredictability and rapidly changing nature of the weather might put people off walking? For example, in Christchurch you can leave the house and be pretty confident that if it’s sunny, it’s relatively unlikely to rain that day. Whereas in Auckland if you leave the house and it’s sunny, unless you’re right in the middle of JAnuary that doesn’t mean much. It could suddenly start raining torrentially just as you leave your office 8 hours later. I certainly get tired sometimes of having to cycle around with both a raincoat, a jumper, sunglasses and suntan cream in my bag. So it might be interesting to look at the relationship between how settled the weather is and how likely people are to walk or cycle. If you had unlimited time that is – or, of course, you could get a social psychologist to take some time out from the crucially important task of trying to work out whether people have more sex in Finland because of the long winters, and work it out for you instead 🙂