Timeline for What kind of curve (or model) should I fit to my percentage data?
Current License: CC BY-SA 4.0
29 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 25, 2023 at 14:51 | comment | added | dipetkov | Thank you so much for responding and congratulations on publishing your work in Nat Sci Reports! I took a look at the paper and I think the interesting structure in the data (what I called "clusters of three") is explained by the design of the experiment and that there were 15 subsamples analyzed. I admit though that I didn't understand lots of the science as I have no knowledge of sequencing. Again, thanks for sharing the extra information! | |
| Aug 25, 2023 at 0:01 | comment | added | teaelleceecee | @dipetkov each data point is an independent measurement as there were three technical replicate samples per viral spike level, so each of these were measured separately, not repeatedly (I hope I interpreted your question correctly). I ended up publishing this work: nature.com/articles/s41598-019-55741-3 Figure 1 illustrates the experiment design, which might help in understanding what the data actually represent and how it was collected. | |
| Aug 20, 2023 at 16:47 | comment | added | dipetkov | 4 years later I have a followup question. (This is the first time I come across this thread.) I notice the observations come in clusters of three with about the same x. Are these data points independent measurements? Or -- and I think this is more likely -- are they replicate measurements taken during the same lab experiment? All the answers so far assume the data consists of 30 independent data points, not 10 independent experiments (5 for each virus) with three repeat measurements each. | |
| Sep 12, 2019 at 13:12 | history | edited | kjetil b halvorsen♦ |
edited tags
|
|
| Jul 26, 2019 at 3:32 | vote | accept | teaelleceecee | ||
| Jul 25, 2019 at 16:02 | history | protected | gung - Reinstate Monica | ||
| Jul 25, 2019 at 15:55 | answer | added | Sam Mason | timeline score: 10 | |
| Jul 23, 2019 at 18:56 | answer | added | Carl Witthoft | timeline score: 4 | |
| Jul 23, 2019 at 13:07 | comment | added | Carl Witthoft | You don't have enough data points in the transition region to claim with any authority that there's a smooth curve. I could just as easily fit a Heaviside function to the points you are showing us. | |
| Jul 23, 2019 at 11:55 | comment | added | jochen | I agree with your supervisor, fitting a sigmoidal curve would be a good choice. The two lines you show in your second plot seem not to be sigmoidal? Have a look on Wikipedia . | |
| Jul 23, 2019 at 6:00 | history | tweeted | twitter.com/StackStats/status/1153545329129734147 | ||
| Jul 22, 2019 at 20:19 | comment | added | James Phillips | After your edit to include the actual data, I have updated my answer with a plot of the 3-parameter logistic type equation using your updated data. | |
| Jul 22, 2019 at 20:12 | comment | added | Ben Bolker |
PS I would encourage you not to suppress standard errors with se=FALSE. Always nice to show people how large the uncertainty actually is ...
|
|
| Jul 22, 2019 at 20:10 | vote | accept | teaelleceecee | ||
| Jul 26, 2019 at 3:32 | |||||
| Jul 22, 2019 at 20:06 | comment | added | teaelleceecee | @BenBolker yes, this removed the extended arch on the curves - thank you! | |
| Jul 22, 2019 at 19:41 | comment | added | Ben Bolker |
try adding method.args=list(family=quasibinomial)) in the arguments to geom_smooth() in your original ggplot code.
|
|
| Jul 22, 2019 at 18:48 | answer | added | Ed V | timeline score: 4 | |
| Jul 22, 2019 at 18:34 | history | became hot network question | |||
| Jul 22, 2019 at 17:48 | answer | added | Nick Cox | timeline score: 14 | |
| Jul 22, 2019 at 17:25 | answer | added | Aksakal | timeline score: 4 | |
| Jul 22, 2019 at 17:14 | history | edited | teaelleceecee | CC BY-SA 4.0 |
Added data
|
| Jul 22, 2019 at 16:33 | history | edited | mkt | CC BY-SA 4.0 |
Edited for clarity
|
| Jul 22, 2019 at 16:20 | answer | added | James Phillips | timeline score: 4 | |
| Jul 22, 2019 at 15:31 | answer | added | mkt | timeline score: 17 | |
| Jul 22, 2019 at 14:28 | comment | added | user158565 | Try (2) piece-wise (linear) model. | |
| Jul 22, 2019 at 11:03 | comment | added | mkt | Seems like a logistic regression would be best, since this is bounded between 0 and 100%. | |
| Jul 22, 2019 at 10:41 | comment | added | Roland | Make use of the "G" in "GAM". I don't know enough about genome coverage (appears to be a percentage?) to recommend a family but you probably want a logit link. | |
| Jul 22, 2019 at 10:30 | review | First posts | |||
| Jul 22, 2019 at 11:29 | |||||
| Jul 22, 2019 at 10:27 | history | asked | teaelleceecee | CC BY-SA 4.0 |