Compute mean and interval across 100 simulations

[netterie]

Title says it all - picking up the conversation from #10 re: the summary table

[netterie]

I added a folder "sims_100" with results and analysis of 100 simulations. You can read in the 100 sim results from the .csv files. Results summarized below and in the html file.

Also:

1. I tried to replicate the 71% statistic for net.comb.2 without luck - which is weird because when I was building the file, I could have sworn I got 71%. After that I consistently got other numbers. See the first 2 sections of the html file. You'll see below that the mean across 100 sims is 62.7%. Oddly, that is one of the few statistic where the number posted online differs noticeably from the mean across 100 sims.
2. The "full-code" section is not totally aligned with the code chunks in SocDistNets.Rmd. It's missing geo.sep calculations for sure.

Largest component - 100 sims

network	largest.Mean	largest.PercOf200	what's online
precov	200.00	100.0	200 (100%)
essl	59.83	29.9	53 (26.5%)
comb.1	181.48	90.7	181 (90.5%)
comb.2	125.43	62.7	142 (71%)

3 degrees of separation - 100 sims

network	geo.3.Mean	geo.3.PercOf200	what's online
precov	199.0	99.5	199 (99.5%)
essl	4.7	2.3	4.2 (2.11%)
comb.1	33.5	16.8	36.8
comb.2	11.9	6.0	12.1

6 degrees of separation - 100 sims

network	geo.6.Mean	geo.6.PercOf200	what's online
precov	199.0	99.5	199 (99.5%)
essl	13.4	6.7	10.1 (5.05%)
comb.1	145.7	72.9	147.2
comb.2	48.5	24.2	52.6

[sgoodreau]

Awesomeness @netterie! We need to add these means into the text, for sure, in some form or other.

My one concern is the timing: a bunch of articles are about to go out citing our current numbers that look at one specific scenario, and the current plots. Even if these numbers are more robust, it'll add confusion if we swap them out at this exact moment. So I see two options:

1. Wait a few days and then rewrite the prose throughout to talk about means and use the new numbers.

2. Add to the prose so that we present numbers for both a specific scenario reflected in the network plots, and also the means across 100 runs. This is scientifically satisfying and solves our media issue above, but I fear it will be information overload for this audience at this stage.

I lean moderately strongly towards #1, but welcome thoughts.

[netterie]

@sgoodreau Sounds good Steve!

[SteveHood]

Steve Hood speaking up. I think this this topic is where our two approaches could be linked.

My shinyapp steve-hood.shinyapps.io/OneFriend allows people to explore not just the senarios you have looked at but others near by and plot out how changes to input parameters change the outputs. I have some text suggesting ways to explore. All of the results are saved in an archive.

My steve-hood.shinyapps.io/OneFriendDataView allows exploration of archive of solutions. When/if there is enough data I'd like to add central tendency and confidence interval graphics to the plots.

I mentioned to Steve G. by email but am recording here so when there is time how to integrate can be considere. I'd be glad to make changes to defaults starting points for the app, and variables stored in my archive if it would be more helpful to you. Currently I have not pushed my sites to a wider audience but I imagine a link to your site would generate more traffic and better idea of the distribution of solutions.

[sgoodreau]

Thanks again @SteveHood Hood. I'm still struggling to find time to look at this in depth, but we have our lab meeting tomorrow and I'll see if others may be able. I agree that linking from us to you would be useful, although now that the press has all passed it does seem like our hits have plummeted back to near-zero. Still worth doing of course.

[SteveHood]

Sorry for the late reply, this message slipped my notice. I welcome corrections to any links back to you on my site. I am getting a modest amount of traffic, probably because I did a bit of Facebook bragging about your site linking to mine!

On Tue, May 19, 2020 at 5:49 PM Steven Goodreau notifications@github.com wrote:

Thanks again @SteveHood https://github.com/SteveHood Hood. I'm still struggling to find time to look at this in depth, but we have our lab meeting tomorrow and I'll see if others may be able. I agree that linking from us to you would be useful, although now that the press has all passed it does seem like our hits have plummeted back to near-zero. Still worth doing of course.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/statnet/COVID-JustOneFriend/issues/14#issuecomment-631168388, or unsubscribe https://github.com/notifications/unsubscribe-auth/APR2VNMM4RP6GVS33Z7H4DLRSMSK7ANCNFSM4MI5MJVQ .