Walkability options for 1000 cities challenge

[carlhiggs]

I've described elsewhere how our current code for spatial indicators is heavily focused on calculation of a walkability index relative to all cities, and that this is problematic now that we seek to generalise our process beyond the preliminary 25 city to 1000+ cities (#136) .

To help start our discussion of this, here are three options --- the first being my recommended one:

1. Present the respective indicators for street connectivity, dwelling density and access to a range of services and amenities ----- but don't calculate the composite score(s)

• pros:
  • very do-able,
  • defers the 'how to do walkability problem' by presenting the core information, and allowing others (perhaps us) to address the 'how' later on;
  • contextually-relevant thresholds could be presented in prose
  • within-city walkability index could still be calculated, if desired
  • the more advanced methods could be incorporated at a later stage
• cons:
  • our current spatial maps and statistics for walkability relative to other cities can't be produced,
  • nor can we say the percentage of population in high walkability neighbourhoods. (but these could be produced as a seperate process however)

2. Identify policy relevant thresholds, and use these as more meaningful benchmarks using essentially the same formula otherwise

• pros:
  • theoretically, this would be a more meaningful walkability index.
  • It would be straightforward to implement (e.g. rather than calculating component z-scores as the standardised difference from the mean value across all neighbourhoods ---ie. (mean-observed)/sd ---, calculate as the difference from policy-relevant target standardised by an agreed tolerance range). Then the standardised components could be summed as before.
• cons:
  • requires review of evidence for thresholds, and a validation study (ie. major blockers when we want people to participate in 1000 cities challenge now)

3. Identify policy relevant thresholds, and use these as more meaningful benchmark and use these in a revised formulation that better accounts for non-linear polarity

(eg more density not necessarily always better) and non-normal distribution (z-score calculation essentially assumes normal distribution, and the implications/influence of non-normal and distinct distributions of component indicators should also be considered for revised method)

• pros:
  • theoretically, this could produce a more meaningful and robust walkability index
• cons:
  • requires exporatory methods study, validation, and review of evidence for thresholds including upper bounds (which evidence is currently lacking on)

What do you think @global-healthy-liveable-cities/ghscic-study-executive ?

[carlhiggs]

Re: option 1 above, I just want to clarify, we can still calculate our traditional 'within city walkability index', its just the 'between city' one that is conceptually challenging because the reference group of cities is unstable shifting. I think option 1 is a good choice, because it is easy to implement, allows us to move forward, and when someone wants to calculate 'between city' walkability they can (just not part of the software workflow, at least for now)

[shiqin-liu]

Agree! I will vote for option 1 as well, with the within-city walkability index, which is more straightforward and fits into our current method pipeline for indicators calculation. Looking at the policy threshold would require much more additional work and exploratory studies, and may be hard to standardize ......

[ecerin]

Option 1 is by far the most feasible. My vote goes to Option 1 as well.

[gboeing]

I think option 1 makes the most sense as well.

[carlhiggs]

Hi @shiqin-liu @ecerin @gboeing --- thanks for the above. We had a discussion with rest of @global-healthy-liveable-cities/ghscic-study-executive this morning and Billie suggested (and others seemed to agree) removing walkability index altogether (but retaining component measures). Later, new measures of walkable neighbourhoods could be added.

Jonathan just e-mailed me with a different proposal, which maybe we should consider:

we could define a benchmark city for comparisons (across cities and across timepoints). If City X is the baseline, we can compare City A with City X, City B with City X and then infer something about City A compared with City B; e.g. (City A has much higher walkability than City X, City B has marginally higher walkability than City X, therefore City A is likely to have higher walkability than City B). ... I was wondering whether it would be possible to generate a synthetic "City X" [to avoid stigmatising any particular city]

I responded with the following:

It's an interesting idea ... and would be easier to implement (eg City X could have the exact same characteristics as the baseline 25 city average, meaning the existing measures would stand, notwithstanding that we'll revise Ghent's)

It doesn't solve other issues though, so i'm not sure if because its easier (ie. than removing walkability altogether) its the best idea (really, don't know what to do; and very focused on phd right now so little time to think/do about this directly!)

• average is by definition a mediocre standard, and likely won't relate to recommendations/goals
• average across diverse cities might result in a standard that isn't meaningful for any of them
• current approach doesn't correct for skewed distributions, and so population density dominates

But I suppose it would allow us to most quickly modify code to get things usable for others; and while this walkability index isn't perfect, no measure is and this one is established in the literature --- we can later supplement it with other measures to consider other aspects

As I indicated, I am not sure what the best approach is -- but Jonathan's suggestion is the easiest to implement so I thought it worth suggesting. The gist is, instead of doing z-scores based on group average, we set the comparison value for each sub-indicator as a value matching the 25-city average from our initial study, representing walkability (in terms of this traditional walkability index) in an 'average city'.

What do you think? I'll try to get the others on here to get their perspectives too.

[mdlowe]

Hi Carl and all,
It's an interesting idea, but do we have any idea about whether the 25 city average is actually a good benchmark for walkability? Or whether it's a good estimate of the average for cities internationally? If we can't be sure, this could be problematic reference point as we expand the number of cities measured.
Melanie

[carlhiggs]

Thanks @mdlowe -- I think we can assume its not a good benchmark, just an 'average' one. The assumption about that is, being above average is 'relatively good', and vice versa. This is inline with how this particular measure has been traditionally calculated and interpreted. I think we agree, there is a place for new additional measures that may be more policy-relevant (ie. a score for meeting specific health and/or policy-related targets).

@global-healthy-liveable-cities/ghscic-study-executive - I need sign off on our approach to walkability still

• Do you think we should
  • remove walkability altogether, retaining sub-indicators, as discussed in our meeting
    • this is somewhat involved, but do-able (changes to processing code and reporting templates)
    • conservative estimate is December for this to be done (as I will be travelling in Oct/Nov)
  • do as Jonathan suggests (retain the 25-city average as a fixed comparator for cities for this walkability index)
    • very easy to implement, I could do it for next week)
  • just remove between-city walkability
    • this is similar to removing walkability altogether, but retains within-city walkability index, so cities could still get a map/data on within city inequities
    • should be able to have done by/in December

[ecerin]

Hi Carl and others, I think it would be useful to postpone the creation of the walkability index (between-city comparison) until we have a good representation of cities across two dimensions: population density and country income level. Best wishes, Ester

[ecerin]

I vote for option 3. Cheers, Ester

[carlhiggs]

Thanks @global-healthy-liveable-cities/ghscic-study-executive; for now I've removed the between city walkability comparisons in the analysis stage, so this issue is at least partially resolved for now. I haven't updated the scorecard reporting for this change, but we can decide on how to approach that later. I didn't remove 'within-city walkability' at this stage (we can later), but I think there was broad concensus on removing between-city walkability so at least that is done.

[VuokkoH]

Hi @carlhiggs and all!

Thanks for sharing this discussion with me.

I have tested the tool for Helsinki, Finland and indeed the between-city comparison does not make sense from our perspective at this stage. I modified the tool a bit like "Option 1" when running the analysis for Helsinki, as I did not download the data for the 25 cities. I'm now planning to expand my tests to other Finnish cities. I can imagine that also others who start to test the tool for their region want to first explore the separate indicators for one city / a set of selected cities. In that sense it's good if the tool is modular from the user's perspective (starting from Option 1), and the city comparison can be the added on later in a meaningful way (~Option 2 / Option 3?).

In addition to global comparisons, I see a need for comparing similar-sized cities, or "regional city buckets" (e.g., across Nordics - regions where similar thresholds make sense).

Thanks again for developing these tools :)

/Vuokko