Brunswick comparison of Base and Cycling exposures

[carlhiggs]

@BelenZapata85 @MahsaAll I have just today been able to run a comparison of exposure results for the base and cycling scenarios; its taken a while to get to this point, so I'm quite excited. These are just for the test region of Brunswick; as per another thread (#168), I have restarted the proper base analysis for Greater Melbourne with a change that should make it faster to complete (I'm cautiously optimistic; its been running for about an hour but its about to get to the walk analysis that previous took prohibitively too long).

The Brunswick analysis (outputs in the Jibe working group Brunswick test folder, under scenOutput/scenario/microData) is just a test --- for speed sake it is done without the full number of iterations (i.e. only 1). This should be fine to get a sense for how things are working I believe --- its a preview, for us to check e.g. are exposures on the correct scale?

I created a rough draft Shiny app that can load a couple of exposure files and compare differences in distribution across variables.

At a glance (all I've done, for now, but thought worth sharing news and starting a conversation), the interesting thing is the reduction in noise (Lden) and air pollution exposures under the cycling scenario:

There are reductions in injuries, in particular for severFatalInjuryWalk --- but we haven't calibrated the injury model at all for Melbourne, so any interpretation of that must be cautious:

But still --- not withstanding the meaning of any absolute values, it is interesting to see relative reductions in fatal injuries under the cycling scenario.

I started this conversation first of all to keep you in the loop @MahsaAll after your work developing the cycling scenario, but also to get your input on the kinds of ranges and distributions of the exposures and outcomes under the two scenarios @BelenZapata85 ---- are these comparable scales to what you are getting for Manchester?

I am a bit concerned about the high levels for noise ---- in particular HA and HSD; extreme right skew there, and I'm not sure its plausible:

Love to hear your thoughts and if there's anything in particular you'd like me to look into on this tomorrow.

[BelenZapata85]

Hi Carl, interesting, thank you. The air pollution exposures look good, tomorrow we can look at the distibutions in numbers. The noise is not a distribution, but we are only interested in the mean value comparison. What about the mmets? lets talk tomorrow, but looking good. I'll take a look at the microdata.

[carlhiggs]

Hi Belen --- great point about the MMET/hours per week ---

There was a small reduction in overall MMET hours for walking (from 3.52 to 4.06; a reduction of 0.54 mmet hours/week), and a notable increase for cycling (from median of 0 to 0.84 mmet hours/week):

Its hard to actually see details of this in the rough box plots -- here are the stats, where df1 is base scenario and df2 is cycling:

Compared to the base scenario, under the cycling scenario the number of persons with cycling mmetHrs/week > 0 in the test suburb increased from 3,109 (13.2%) to 15,073 (64.2%). While most persons (86.8%) had 0 cycling mmet hours/week under the base scenario (mean 0.4 hours), under the cycling scenario the median [IQR] was 0.8 [0.0, 2.3] mmet hours/week (mean 1.6 hours)

> (df1$mmetHr_cycle > 0) %>% table()
.
FALSE  TRUE
20360  3109

> (df2$mmetHr_cycle > 0) %>% table()
.
FALSE  TRUE
 8396 15073

> (df1$mmetHr_cycle > 0) %>% table() %>% prop.table()
.
    FALSE      TRUE
0.8675274 0.1324726

> (df2$mmetHr_cycle > 0) %>% table() %>% prop.table()
.
    FALSE      TRUE
0.3577485 0.6422515

> df1$mmetHr_cycle %>% summary()
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.0000  0.0000  0.3583  0.0000 23.7215

> df2$mmetHr_cycle %>% summary()
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.0000  0.0000  0.8354  1.5985  2.3325 23.7215

There was a small increase in severe fatal injury risk for cycling (makes sense, given the shift), however this was more than offset by a notable decrease in severe fatal injury from walking. The latter presumably relates to the induced shift from driving to cycling; fewer cars on the road. Compared to the base scenario, under the cycling scenario the number of cyclists in the test suburb increased from 3,109 (13.2%) to 15,073 (64.2%).