tsset to use with itsa

[Owen Gallupe]

Hi,

I'm getting stuck on what I suspect is a small point and couldn't find the solution. I would like to conduct a single-group interrupted time series analysis using Linden's itsa package using the data copied at bottom.

The intervention is at time point 31.

tsset id obsday
itsa sleep, single trperiod(31) lag(1) posttrend figure

This generates the following error:

d(31) invalid
r(198);

Clearly my knowledge of time series data is lacking, but I can't seem to get this to work.

I'm using 'itsa' from SSC in Stata 17.0.

Thanks.

Owen

* Example generated by -dataex-. For more info, type help dataex
clear
input float(id group obsday sleep)
 1 1  1 353
 1 1  9 345
 1 1 15 340
 1 1 21 337
 1 1 26 324
 1 1 33 332
 1 1 41 326
 1 1 49 329
 2 1  1 362
 2 1  9 361
 2 1 15 355
 2 1 23 365
 2 1 28 350
 2 1 33 343
 2 1 39 333
 2 1 45 320
 3 1  1 345
 3 1  9 360
 3 1 16 347
 3 1 24 348
 3 1 29 351
 3 1 37 348
 3 1 44 334
 3 1 50 335
 4 1  1 338
 4 1  9 357
 4 1 16 384
 4 1 21 418
 4 1 26 408
 4 1 31 404
 4 1 36 423
 4 1 43 403
 5 1  1 341
 5 1  8 341
 5 1 15 356
 5 1 21 368
 5 1 27 377
 5 1 33 367
 5 1 40 363
 5 1 46 362
 6 1  1 353
 6 1  8 340
 6 1 15 334
 6 1 22 330
 6 1 28 327
 6 1 35 306
 6 1 41 321
 6 1 49 319
 7 1  1 373
 7 1  7 373
 7 1 13 380
 7 1 21 377
 7 1 27 372
 7 1 31 372
 7 1 38 361
 7 1 45 347
 8 1  1 323
 8 1  7 327
 8 1 12 335
 8 1 19 307
 8 1 25 314
 8 1 32 284
 8 1 40 244
 8 1 47 235
 9 1  1 353
 9 1  7 361
 9 1 13 350
 9 1 19 342
 9 1 25 317
 9 1 30 301
 9 1 36 306
 9 1 43 303
10 1  1 365
10 1  8 366
10 1 14 368
10 1 21 374
10 1 27 396
10 1 34 369
10 1 39 393
10 1 47 391
11 1  1 343
11 1  7 347
11 1 14 334
11 1 20 327
11 1 25 326
11 1 33 312
11 1 38 348
11 1 45 356
12 1  1 360
12 1  8 356
12 1 14 342
12 1 20 360
12 1 26 366
12 1 32 350
12 1 38 383
12 1 45 385
13 1  1 347
13 1  8 353
13 1 12 339
13 1 18 316
end
format %td obsday
label values group group
label def group 1 "Control", modify

[Rich Goldstein]

based on the help file, which says, "Dates should be specified using the respective pseudofunction (see datetime), such as trperiod(2020) for a four-digit year or trperiod(2019m11) for quarterly data." try

itsa sleep, single trperiod(31jan1960) lag(1) posttrend figure

however, since you have more than one panel, you also need to specify the "treatid" option - see

h itsa

[Owen Gallupe]

Thank you, Rich! I clearly was missing something quite simple in my time series specification.

The second part is another area where I think I'm off the track a little. I managed to follow other guidance to conduct an ITSA using this code:

mkspline knot1 31 knot2 = obsday, marginal

*First, generate an intervention indicator (0=pre-intervention, 1=intervention).
gen period = .
replace period = 0 if obsday < 31
replace period = 1 if obsday >= 31 & !missing(obsday)

*Estimate ITSA
xtmixed sleep i.period c.knot1 c.knot2 || id: period knot1 knot2, covariance(unstructured)

*Test different in slopes before vs after intervention
margins, expression(_b[knot1] + _b[knot2])

I try to replicate it using 'itsa' package which would enable the use of Newey-West standard errors and the use of the 'actest' command to test the autocorrelation structure. I would have thought the following would work, but not so much.

tsset id obsday, daily
itsa sleep, single trperiod(31jan1960) lag(1) posttrend figure

But with multiple respondents, I get the message about having to specify a 'treatid' option. Is there a way to specify either 'tsset' or 'itsa' to include all respondents in the same analysis?

Or am I possibly just misunderstanding what the 'itsa' can do?

Thanks.

Owen

[Rich Goldstein]

I'm confused about what you are trying to do; however, yes, if you only include a timevar in your -tsset- statement you should get something

[Owen Gallupe]

I'm not being very clear. Let me try again in a more general way

I've got the data above and I would like to conduct a single-group interrupted time series analysis using the 'itsa' package to see if an intervention at day 31 has any effect. I'm getting stuck on implementing it.

Thanks!

Owen