0

Yet another data scientist not keeping a safe distance from COVID-19 data.

I am making a plot of the infection and deahts doubling times which I am calculating on a running basis, say previous 14 days. I use the glm function in R to perform an log-link regression to get the doubling time and confidence interval (high/low) for that value. I'm putting these in the mapping as ymin and ymax. I get the confidence band, but it is understandably jagged as is the data. Is there a simple way to smooth the confidence bands?

covid_infection_folding %>%
  ggplot() + 
  geom_point(aes(x=Date, y=US_Infections, color="US Infections")) +
  geom_point(aes(x=Date, y=US_Deaths, color="US Deaths")) +
  geom_smooth(
    data=covid_infection_folding,
    mapping=aes(x=Date, y=US_Infections, ymin=US_Infections_low, ymax=US_Infections_high, color="US Infections"),
    stat="identity"
    ) +
  geom_smooth(
    data=covid_infection_folding,
    mapping=aes(x=Date, y=US_Deaths, ymin=US_Deaths_low, ymax=US_Deaths_high, color="US Deaths"),
    stat="identity"
    ) + 
  labs(
    y="US Covid-19 Doubling Time (Days)",
    title="Doubling Time (95% confidence intervals)",
    subtitle="Based on purevious 14 days"
  )

covid_infections_folding.csv

"Date","US_Infections","US_Infections_low","US_Infections_high","World_Infections","World_Infections_low","World_Infections_high","US_Deaths","US_Deaths_low","US_Deaths_high","World_Deaths","World_Deaths_low","World_Deaths_high"
2020-03-14,2.49983739883223,2.38168561730312,2.62426904052848,15.770263051682,13.9876095552037,18.0489074832855,3.99043275230409,3.69236832976168,4.32140858392337,12.8989698838882,11.4099582259867,14.8038432131489
2020-03-15,2.588057654306,2.47627312530032,2.70530124580819,14.1052241582214,12.5458193107193,16.0815164642284,4.18843478882557,3.99109724055167,4.39934762358077,11.4155530248942,10.1442558270398,13.0197551751998
2020-03-16,2.59916192635231,2.51169240835164,2.68989407531561,12.7183934752834,11.3821909647131,14.3853177366056,3.92980192885667,3.67451528683076,4.20867530268657,10.1892563438137,9.13874885076098,11.4843983540597
2020-03-17,2.50209502182501,2.41277807296645,2.59484269035652,11.5822559952822,10.4704174205308,12.9369688457438,3.65991741725404,3.38578333892515,3.96218180909321,9.22152165144878,8.36609944993533,10.2483894056885
2020-03-18,2.61215755183853,2.5110308935239,2.71767102078119,10.6653403951574,9.74603324802977,11.7586509149994,3.7539849058838,3.50894146518389,4.0213997894202,8.50182769556017,7.82837103723028,9.28511058399383
2020-03-19,2.12103811052124,1.86668177353069,2.40773667867803,9.68220899681998,8.85264088937122,10.664710714973,2.99387387850563,2.54689938108029,3.52708155887181,7.85987840683811,7.30624027338352,8.49070096408631
2020-03-20,1.99079441567267,1.81410759357454,2.18405773648449,8.80015137195741,8.06816675831183,9.65957369382066,2.80956713501713,2.47874770062383,3.18862071502173,7.25673080882034,6.76600476407113,7.81116618913957
2020-03-21,2.05278151635975,1.92120912686599,2.19340371313652,8.08237805494367,7.46494738129346,8.79499183281471,2.7798252180882,2.53323277524509,3.0530722207346,6.73383124186039,6.27979418641987,7.2451124137753
2020-03-22,2.17252528926468,2.04949059350162,2.30316218551139,7.58160689577727,7.09323516620424,8.13086338523261,2.65984002237084,2.47418817905869,2.86058591174651,6.35099806759453,5.97729955890051,6.76431384218689
2020-03-23,2.27085728917173,2.15744423749247,2.39046362864804,7.18006161242301,6.78634277949618,7.61413540189244,2.56825970738847,2.42667381891871,2.71864717722737,6.099512763381,5.8048741730648,6.41894911090602
2020-03-24,2.4531812109224,2.30077480144939,2.6164374199738,6.93254739983745,6.63082808556837,7.25805297994967,2.59389681397186,2.48718548076248,2.70554268775727,5.92711453226658,5.70556072017153,6.16268318967908

Note: the raw data is from here: https://github.com/CSSEGISandData/COVID-19

abalter
  • 9,663
  • 17
  • 90
  • 145
  • Are you asking for a statistical recommendation or a purely aesthetic one? Typically to get smoothed data you need to make a modeling assumption. I don't think this is really a specific programming question that's appropriate for Stack Overflow. If you need help modeling data or asking for plot recommendations, you should ask at [stats.se] where statistical questions are on topic. At the very least make the data in your question itself [reproducible](https://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example) so it doesn't rely on data hosted externally. – MrFlick Mar 27 '20 at 22:31
  • Hi @MrFlick. Your question is a good one. It's aesthetic. The link just shows where I got the raw data. The plotted data is pasted into the post (It takes some bit of processing). I think I'm ok with the modeling--I stated how I performed the exponential fit. Now, if there is a statistical method that would, say, allow me to infer higher resolution data or something that would make the data smoother, I'm open. Otherwise, it's a `ggplot2` question. I can let `geom_smooth` do a loess fit (the default) but then I don't get real confidence intervals (based on fitting the data). – abalter Mar 27 '20 at 22:45

0 Answers0