Timeline for Is this an appropriate method to test for seasonal effects in suicide count data?
Current License: CC BY-SA 3.0
33 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 24, 2015 at 19:42 | vote | accept | svannoy | ||
| Apr 24, 2015 at 19:40 | vote | accept | svannoy | ||
| Apr 24, 2015 at 19:42 | |||||
| Apr 24, 2015 at 19:40 | vote | accept | svannoy | ||
| Apr 24, 2015 at 19:40 | |||||
| Apr 24, 2015 at 19:40 | vote | accept | svannoy | ||
| Apr 24, 2015 at 19:40 | |||||
| Apr 18, 2015 at 1:54 | comment | added | svannoy | I like the plot, I think I should have included something like it initially anyhow. I was looking at histograms to get a sense of trend/season, but this I think is better. | |
| Apr 17, 2015 at 1:50 | comment | added | rnso | I always thought this plot was useful but @Tim did not like it. I have undeleted my answer. | |
| Apr 17, 2015 at 1:38 | comment | added | Glen_b | svannoy -- I took the liberty of moving rnso's plot to your Q. @rnso -- I hope you don't mind. If you have some strong objection, let me know. | |
| Apr 17, 2015 at 1:36 | comment | added | Glen_b | forecaster -- because in count data spread is related to mean. e.g. consider a Poisson count $y_t$, with mean $\mu_t$. In that case, $\text{Var}(y_t)=\mu_t$, so as the mean increases, the variance (and so also the spread) increases. [Indeed, often you find spread to be related to mean in some way for almost any data that has an upper or lower bound; there are clear reasons why this might be expected in those cases.] | |
| Apr 17, 2015 at 1:32 | history | edited | Glen_b | CC BY-SA 3.0 |
added 294 characters in body
|
| Apr 17, 2015 at 0:34 | comment | added | forecaster | @Glen_b, yes! the data requires transformations as it can be seen as the level shifts in 2009, there is more fluctuations/variability. Can you please let me know how and why this happens in count data vis-a-vis time series ? | |
| Apr 16, 2015 at 20:34 | history | edited | svannoy | CC BY-SA 3.0 |
I've updated the question to provide new data that people can use for analyses. Specifically, I've added results of a modification in analytic technique and I've provided data to model season by the number of daylight minutes throughout the year
|
| Apr 15, 2015 at 4:26 | comment | added | Elvis | Wow, thanks. I’ll try to rerun the Poisson regression with the daily data! | |
| Apr 14, 2015 at 13:54 | comment | added | Glen_b | Since your data are counts, I'd expect variance to be related to the mean. The usual time series models don't account for that (however, you might try say a transformation, perhaps a Freeman-Tukey, say), or you could look at a time series model that's designed for count data. (If you don't do this it may not be a huge problem since the number only ranges over a factor of two or so.) | |
| Apr 14, 2015 at 12:27 | comment | added | svannoy | @Elvis - I've posted a link to the daily count data. The data comes from death certificates which are 'public record' but require a process to obtain; however, the aggregated count data does not. PS - I tried the link myself and it worked, but I've not posted to a public dropbox folder in this way before so please let me know if the link does not work. | |
| Apr 14, 2015 at 12:25 | comment | added | svannoy | Thanks @NickCox for cleaning up the grammar and putting in the links. | |
| Apr 14, 2015 at 12:23 | history | edited | svannoy | CC BY-SA 3.0 |
I added a link to a csv file with the daily suicide counts
|
| Apr 12, 2015 at 21:00 | history | edited | Nick Cox | CC BY-SA 3.0 |
added 2 characters in body
|
| Apr 12, 2015 at 19:15 | answer | added | forecaster | timeline score: 7 | |
| Apr 11, 2015 at 3:45 | answer | added | rnso | timeline score: 1 | |
| Apr 9, 2015 at 21:28 | history | edited | svannoy | CC BY-SA 3.0 |
deleted 4 characters in body
|
| Apr 9, 2015 at 7:22 | comment | added | Elvis | Is that from a public dataset? Could you make the week-by-week or even day-by-day data available? | |
| Apr 8, 2015 at 8:26 | answer | added | Elvis | timeline score: 13 | |
| Apr 7, 2015 at 16:53 | comment | added | Nick Cox | The wording "one of our 50 states" implies that all readers belong to the United States. Manifestly many aliens lurk here too. | |
| Apr 7, 2015 at 13:35 | history | edited | Silverfish |
add some tags
|
|
| Apr 7, 2015 at 12:59 | answer | added | javlacalle | timeline score: 8 | |
| Apr 5, 2015 at 0:09 | comment | added | svannoy | @forecaster - the data come from death certificate data at the (a) state level. I've "dumped" the time series, the data itself I believe is too large to go into the post. | |
| Apr 5, 2015 at 0:07 | history | edited | svannoy | CC BY-SA 3.0 |
I added several items requested (plot of daily counts) and one's I thought would be helpful (table of time series data, plot of residuals, details of the stationarity tests), and updated my understanding thus far.
|
| Apr 4, 2015 at 12:36 | comment | added | svannoy | Thanks Richard, I was basing my conclusion more on the fact that nsdiffs() returned zero rather than actually interpreting the model. I need to go back and more fully understand the model that was selected by auto.arima() | |
| Apr 4, 2015 at 9:53 | comment | added | Richard Hardy |
In the SARIMA model there are the seasonal sar1 and sma1 terms, so your statement based on the model selected, there is a trend but not a seasonal component does not seem right. The sar1 and sma1 make up the seasonal component.
|
|
| Apr 4, 2015 at 0:17 | comment | added | rnso | There seem to be a trend peaking at May-July but whether it is statistically significant is the question. It may be better if you plot by day of the year (1-366) rather than plotting monthly values. It may help if you post here 366 numbers indicating number of deaths on each day of the year for entire data. | |
| Apr 4, 2015 at 0:08 | comment | added | forecaster | Very interesting problem Can you please post the data and also please share the source for this data set? | |
| Apr 3, 2015 at 22:51 | review | First posts | |||
| Apr 3, 2015 at 23:07 | |||||
| Apr 3, 2015 at 22:47 | history | asked | svannoy | CC BY-SA 3.0 |