Hi all, I have a dataset that includes rows with - people who viewed TV advertisements at a given time, and whether or not these people visited the website of the advertiser afterwards (+ the timestamp of this visit) - a match for each of these people who did NOT see the TV advertisement at this same time, and whether or not these people visited the website of the advertiser afterwards (+ the timestamp of this visit) To give a simplified idea: On each row we also have a bunch of dummy variables for the week, weekend vs weekday, daypart (primetime etc). I use PROC PHREG to model the duration of the treatment effect (=seeing a TV ad) on online response (web visit). We included a random term for the advertisement, because the same advertisement is shown many times on different timestamps and to different people. We would like to produce a hazard graph including the time-related control variables + including the random term. Is that possible? Ideally, I'd get something similar to the below output that I generated with PROC LIFETEST, but including control variables and the random term (which is not possible with PROC LIFETEST). PROC PHREG only produces a cumulative hazard graph and a survival graph. By using the COVS statement, I can control for the time-related variables. For now, I calculated the hazard rates by doing cumhaz - lag(cumhaz) directly in the output dataset. Not sure if this is allowed? Also, I cannot produce any output results if I include the random statement (The PLOTS=, OUTPUT=, ... options are ignored for a frailty model analysis). Does anyone know of a method to go around this? Thank you so much in advance for any input! current code below: DATA covs; FORMAT treatment best12. d_daytime_9_16 best12. d_earlyfringe_16_20 best12. d_primetime_20_23 best12. d_latefringe_23_5 best12. d_weekend best12. d_week14 best12. d_week15 best12. d_week16 best12. d_week17 best12. d_week18 best12. d_week19 best12. d_week20 best12. d_week21 best12. d_week22 best12. d_week23 best12. d_week24 best12. d_week25 best12. d_week26 best12.; INPUT treatment d_daytime_9_16 d_earlyfringe_16_20 d_primetime_20_23 d_latefringe_23_5 d_weekend d_week14 d_week15 d_week16 d_week17 d_week18 d_week19 d_week20 d_week21 d_week22 d_week23 d_week24 d_week25 d_week26 ; DATALINES; 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 ; RUN; ods graphics on; PROC PHREG DATA=metrics_v4 PLOTS(overlay range=0,10080)=(survival cumhaz); CLASS advertisement treatment(ref="0") visit(ref="0") d_daytime_9_16(ref="0") d_earlyfringe_16_20(ref="0") d_primetime_20_23(ref="0") d_latefringe_23_5(ref="0") d_weekend(ref="0") d_week14(ref="0") d_week15(ref="0") d_week16(ref="0") d_week17(ref="0") d_week18(ref="0") d_week19(ref="0") d_week20(ref="0") d_week21(ref="0") d_week22(ref="0") d_week23(ref="0") d_week24(ref="0") d_week25(ref="0") d_week26(ref="0") ; MODEL survivaltime*visit(0) = treatment d_daytime_9_16 d_earlyfringe_16_20 d_primetime_20_23 d_latefringe_23_5 d_weekend d_week14 d_week15 d_week16 d_week17 d_week18 d_week19 d_week20 d_week21 d_week22 d_week23 d_week24 d_week25 d_week26 ; BASELINE COVARIATES=covs OUT=base SURVIVAL=surv CUMHAZ=cumhaz / ROWID=treatment; RUN; DATA test_haz; SET base; haz = cumhaz - lag(cumhaz); if cumhaz=0 then haz = .; RUN;
... View more