-
Notifications
You must be signed in to change notification settings - Fork 0
/
Replisims Juan and Felix.Rmd
662 lines (532 loc) · 33.8 KB
/
Replisims Juan and Felix.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
---
title: Replication Report @Brookhart2006
author1: Juan Claramunt\textsuperscript{1},
author2: Felix Clouth\textsuperscript{2}
organization1: Leiden University
organization2: Tilburg University
date: \today
params:
iblue: 008080
igray: d4dbde
documentclass: article
fontsize: 10
papersize: a4paper
bibliography : ["My Collection.bib"]
output:
RepliSimReport:::replisim_report:
keep_tex: TRUE
latex_engine: xelatex
resetStyleFiles: FALSE
header-includes:
- \newcommand{\iblue}{`r params$iblue`}
- \newcommand{\igray}{`r params$igray`}
include-before:
- \renewcommand{\contentsname}{Table of Contents}
- \renewcommand{\pagename}{Page}
editor_options:
markdown:
wrap: 72
---
```{r setup, include = FALSE}
# packages
library(dplyr)
library(knitr)
library(xtable)
load("ResultsEnvironment.RData")
# settings
knitr::opts_chunk$set(echo = FALSE, message = FALSE, warning = FALSE)
```
\urlstyle{same}
<!--ensure urls have the same style as the rest of the text-->
\maketitle
Replication Report, Brookhart et al. (2006)
\subsection*{Abstract}
In this document, we describe a replication study of "Variable Selection
for Propensity Score Models" by Brookhart et al. (Brookhart et al.
2006). This report includes an analysis of the data-generating
mechanism, simulation study set-up and methods described in the original
paper, the feasibility of its implementation based on the information
available in the original document and the results we were able to
replicate. Furthermore, we analyze whether the presentation of the
methods and results on the original paper can be improved. We were able to replicate the results of the first experiment of the original paper by making a few small assumptions. However, we were not able to replicate the results of experiment two. This was probably caused by a lack of explanation of certain concepts used on the article, the lack of experience of the authors of this report and the extra difficulty that requires replicating a figure instead of a table.
\vskip 2em
\noindent\makebox[\textwidth]{\large Correspondence concerning this replication report should be addressed to:}
\par
\noindent\makebox[\textwidth]{\large f.j.clouth@tilburguniversity.edu}
\par
\clearpage
\section{Introduction}
This replication report documents the replication attempt of the
simulation study @Brookhart2006. Following the definition of
@rougier_sustainable_2017-1 we understand the replication of a published
study as writing and running new code based on the description provided
in the original publication with the aim of obtaining the same results.
This replication study was independently done by another pair of
researchers. In the discussion section we will include an overview of
the common findings an differences.
\section{Method}
\subsection{Information basis}
This replication was conducted mainly on information provided in
@Brookhart2006. Software code or additional information (such as
supplementary materials) were not available. The addition of the open
source code (according to the original paper, it exists but it is not
published) will make the replication of the study much easier.
Additionally, we used information provided in the splines2 and ROCR
R-package documentation. However, this information was not referred to
in the main article.
\subsection{Data Generating Mechanism}
Information provided in the above mentioned sources indicated that the
following simulation factors were systematically varied in generating
the artificial data.
\subsubsection{Experiment 11}
+-------------+---+----------------------------------------------------+
| Simulation | N | Levels |
| factor | o | |
| | . | |
| | l | |
| | e | |
| | v | |
| | e | |
| | l | |
| | s | |
+=============+===+====================================================+
| *Varied* | | |
+-------------+---+----------------------------------------------------+
| Sample size | 2 | 500, 2500 |
+-------------+---+----------------------------------------------------+
| *Fixed* | | |
+-------------+---+----------------------------------------------------+
| | | 0.5 |
| $\alpha_{0}$| | |
+-------------+---+----------------------------------------------------+
| | | 4 |
| $\alpha_{1}$| | |
+-------------+---+----------------------------------------------------+
| | | 1 |
| $\alpha_{2}$| | |
+-------------+---+----------------------------------------------------+
| | | 0 |
| $\alpha_{3}$| | |
+-------------+---+----------------------------------------------------+
| | | 0.5 |
| $\alpha_{4}$| | |
+-------------+---+----------------------------------------------------+
| $\beta_{0}$ | | 0 |
+-------------+---+----------------------------------------------------+
| $\beta_{1}$ | | 0.5 |
+-------------+---+----------------------------------------------------+
| $\beta_{2}$ | | 0 |
+-------------+---+----------------------------------------------------+
| $\beta_{3}$ | | 0.75 |
+-------------+---+----------------------------------------------------+
| Dichotomous | | sampled from rbern(N, |
| exposure A | | prob=pnorm($q=\beta_{0} + \beta_{1}*X_{1}+\beta_{2}*X_{2}+\beta_{3}*X_{3}$))|
| | | |
+-------------+---+----------------------------------------------------+
| Outcome Y | | sampled from rpois(N, lamda=$exp(\alpha_{0}+\alpha_{1}*(((1+exp(-3*X_{1}))^{-1})-0.5)+\alpha_{2}*X_{2}+\alpha_{3}*X_{3}+\alpha_{4}*A))$|
+-------------+---+----------------------------------------------------+
| Three | | independently sampled from rnorm(N, 0, 1) |
| covariates | | |
+-------------+---+----------------------------------------------------+
\subsubsection{Experiment 1 (sensitivity analysis)}
+-------------+---+----------------------------------------------------+
| Simulation | N | Levels |
| factor | o | |
| | . | |
| | l | |
| | e | |
| | v | |
| | e | |
| | l | |
| | s | |
+=============+===+====================================================+
| *Varied* | | |
+-------------+---+----------------------------------------------------+
| Standard | 2 | 0.5, 1.5 |
| deviation | | |
| of | | |
| covariates | | |
+-------------+---+----------------------------------------------------+
| | 3 | 0.25, 0.5, 1 |
| $\alpha_{4}$| | |
+-------------+---+----------------------------------------------------+
| $\beta_{0}$ | 2 | -1, 0 |
+-------------+---+----------------------------------------------------+
| *Fixed* | | |
+-------------+---+----------------------------------------------------+
| Sample size | | 500 |
+-------------+---+----------------------------------------------------+
| | | 0.5 |
| $\alpha_{0}$| | |
+-------------+---+----------------------------------------------------+
| | | 4 |
| $\alpha_{1}$| | |
+-------------+---+----------------------------------------------------+
| | | 1 |
| $\alpha_{2}$| | |
+-------------+---+----------------------------------------------------+
| | | 0 |
| $\alpha_{3}$| | |
+-------------+---+----------------------------------------------------+
| $\beta_{1}$ | | 0.5 |
+-------------+---+----------------------------------------------------+
| $\beta_{2}$ | | 0 |
+-------------+---+----------------------------------------------------+
| $\beta_{3}$ | | 0.75 |
+-------------+---+----------------------------------------------------+
| Dichotomous | | sampled from rbern(N, |
| exposure A | | prob=pnorm($q=\beta_{0} + \beta_{1}*X_{1}+\beta_{2}*X_{2}+\beta_{3}*X_{3}$)) |
+-------------+---+----------------------------------------------------+
| Outcome Y | | sampled from rpois(N, lamda=$exp(\alpha_{0}+*\alpha_{1}*(((1+exp(-3*X_{1}))^{-1})-0.5)+\alpha_{2}*X_{2}+\alpha_{3}*X_{3}+\alpha_{4}*A))$|
+-------------+---+----------------------------------------------------+
\subsubsection{Experiment 2}
+-------------+---+----------------------------------------------------+
| Simulation | N | Levels |
| factor | o | |
| | . | |
| | l | |
| | e | |
| | v | |
| | e | |
| | l | |
| | s | |
+=============+===+====================================================+
| *Varied* | | |
+-------------+---+----------------------------------------------------+
| Sample size | 2 | 500, 2500 |
+-------------+---+----------------------------------------------------+
| | 2 | [0, .01, ..., .2] |
| $\alpha_{1}$| 1 | |
+-------------+---+----------------------------------------------------+
| $\beta_{1}$ | 2 | [0, .05, ..., 1.25] |
| | 6 | |
+-------------+---+----------------------------------------------------+
| *Fixed* | | |
+-------------+---+----------------------------------------------------+
| | | 0.5 |
| $\alpha_{0}$| | |
+-------------+---+----------------------------------------------------+
| | | 1 |
| $\alpha_{2}$| | |
+-------------+---+----------------------------------------------------+
| | | 0 |
| $\alpha_{3}$| | |
+-------------+---+----------------------------------------------------+
| | | 0.5 |
| $\alpha_{4}$| | |
+-------------+---+----------------------------------------------------+
| $\beta_{0}$ | | 0 |
+-------------+---+----------------------------------------------------+
| $\beta_{2}$ | | 0 |
+-------------+---+----------------------------------------------------+
| $\beta_{3}$ | | 0.75 |
+-------------+---+----------------------------------------------------+
| Three | | independently sampled from rnorm(N, 0, 1) |
| covariates | | |
+-------------+---+----------------------------------------------------+
| Outcome Y | | sampled from |
| | | rpois(exp($\alpha_{0}+\alpha_{1}*X_{1 }+\alpha_{2}*X_{2}+\alpha_{3}*X_{3}+\alpha_{4}*A$)) |
| | | |
+-------------+---+----------------------------------------------------+
| Dichotomous | | sampled from rbern(N, |
| exposure A | | prob=pnorm($q=\beta_{0} + \beta_{1}*X_{1}+\beta_{2}*X_{2}+\beta_{3}*X_{3}$)) |
+-------------+---+----------------------------------------------------+
The design was full-factorial in the main analyses of Experiment 1 and
Experiment 2. For the sensitivity analyses of Experiment 1, all
parameters were held at their default values while a single parameter
was altered.
\subsection{Research Comparison}
The study compares different specifications of a propensity scores (PS)
model using a Monte-Carlo simulation. The model examines the effects of
the inclusion of three types of covariates: X1 (a true confounder), X2
(a predictor of the outcome) or X3 (a variable related to the exposure
but not the outcome).
The exposure effects are estimated using two methods: (1) by adjusting
for the PS in a multivariate outcome model in which the effect of the
estimated PS was modeled through a regression spline. In this case, the
propensity score model fit is given by the following formula:
$$
E[Y|\hat{PS},A] = exp \{ \lambda + \sum_k \psi_k B_k(\hat{PS}) + \gamma A \}
$$
where $\lambda$ is the baseline rate, $B_k$ are B-spline basis
functions and $\gamma$ is the treatment effect. (2) Using
subclassification in which strata are defined by quintiles of the
estimated propensity scores and taking the average treatment effect
across strata.
\subsection{Performance measures}
To measure the performance of the different specifications of the PS
models of the original simulation study, the authors measured the bias,
the variance and the mean square error of the estimates of the parameter
$\alpha_{4}$ and the average C-stat.
\subsection{Technical implementation}
While the original simulation study was carried out in R version 1.9.1
running on a Windows XP platform, our replication was implemented using
the R programming environment version 4.0.3 on a Windows 10 environment
(details regarding software versions can be obtained from the section
Reproducibility Information). The corresponding R code can be obtained
from <https://github.com/replisims/your_repo>.
<!-- Add zenodo doi once obtained MISSING-->
The following table provides an overview of replicator degrees of
freedom, i.e. decisions that had to be made by the replicators because
of insufficient or contradicting information. Issues were resolved by
discussion among the replicators. Decisions were based on what the
replicators perceived to be the most likely implementation with
likeliness estimated by common practice and/or guideline
recommendations. Wherever feasible multiple interpretations where
implemented.
+-------------------------+-------------------------+------------------+
| Issue | Replicator decision | Justification |
+=========================+=========================+==================+
| It was not clear | Estimate the PS | It should result |
| whether to estimate the | | in a better fit |
| PS or to calculate them | | for each |
| from the population | | simulated |
| model | | dataset. |
+-------------------------+-------------------------+------------------+
| The implementation of | We used the function | Only available |
| the cubic spline | bs() in splines | function in R. |
| estimation was not | | We used the |
| clearly explained. | package, with | default options. |
| Moreover, the reference | specifying three | |
| included does not | | |
| provide enough insight | knots splines::bs(PS, | |
| | knots = | |
| | | |
| | quantile(PS, probs = | |
| | c(0.25, 0.5, | |
| | | |
| | 0.75))) | |
+-------------------------+-------------------------+------------------+
| Subclassification | We used the mean as | It is the most |
| approach not fully | estimator | reasonable |
| defined. Not clear how | | choice. |
| to obatin the estimates | | |
| of the exposure rate | | |
| within strata. | | |
+-------------------------+-------------------------+------------------+
| No description of | Hmisc::somers2(PS, | Most likely |
| c-statistic | | meaning of the |
| | data[,"exposure"])["C"] | c-statistic. |
| | | Most common R |
| | | package to |
| | | compute the area |
| | | under the ROC |
| | | curve. |
+-------------------------+-------------------------+------------------+
| Some values required | Used the same values we | If the values |
| for experiment 2 were | used in experiment 1 | are omitted, the |
| missing | | most likely |
| | | reason is that |
| | | the values are |
| | | the same for |
| | | both |
| | | experiments. |
+-------------------------+-------------------------+------------------+
| Definition and scales | No solution | |
| of the variables | | |
| measured in the axes of | | |
| the figures of | | |
| experiment 2 were not | | |
| clearly described. | | |
+-------------------------+-------------------------+------------------+
| Estimators of | Use the estimators of | It is the most |
| experiment 2 were not | experiment 1 that | reasonable |
| clearly explained. | fitted better the brief | choice. |
| | explanation | |
+-------------------------+-------------------------+------------------+
| According to the | No solution | |
| Methods section, the PS | | |
| was adjusted using the | | |
| spline function in | | |
| experiment 2. However, | | |
| in the results section | | |
| we can find the | | |
| quintile approach as | | |
| well. | | |
+-------------------------+-------------------------+------------------+
\subsection{C-statistic issue}
In a first version of the original paper we were trying to replicate, it
was not specified what the *c* statistic mean. Therefore, it was not
clear what measure we were replicating nor how we were supposed to do
so. In latter versions, it is stated that the *c*-statistics they use in
this research is the area under the receiver operating characteristic
(ROC) curve.
\subsection{Experiment 2 issue}
Whilw trying to replicate the results of the second experiment we found
multiple issues. First, they use multiple transformation of the results
that are not defined in the paper. Second, it is far more difficult to compare the results obtained in a graph than in a table. While we could easily check that the results we obtained for experiment 1 were exactly the same as the results obtained in the original paper, it was incredibly difficult to do the same with experiment 2. Furthermore, there is no open code available to replicate these figures. These difficulties did not allow us to replicate the figures of experiment 2.
\section{Results}
\subsection{Experiment 1}
In this section we show the tables we obtain during the replication of experiment 1.
```{=tex}
\begin{table}
\caption{Table 1. Simulation experiment 1, with results based on an analysis in which the propensity score is entered into an outcome model as a parametric spline term.}
\begin{tabular}{l r r r r r r r r}
\hline
\multicolumn{8}{c}{Variable(s) in propensity score model}\\
\hline
& $X_{1}$ & $X_{2}$ & $X_{3}$ & $X_{1}$, $X_{2}$ & $X_{1}$, $X_{3}$ & $X_{2}$, $X_{3}$ & $X_{1}$, $X_{2}$, $X_{3}$ & None\\
\hline
n = 500\\
Bias & 0.008 & 0.601 & 0.744 & 0.006 & 0.007 & 0.744 & 0.006 & 0.599\\
Variance & 0.030 & 0.025 & 0.047 & 0.020 & 0.040 & 0.038 & 0.030 & 0.040\\
MSE & 0.030 & 0.386 & 0.600 & 0.020 & 0.040 & 0.591 & 0.030 & 0.398\\
avg c-stat & 0.67 & 0.52 & 0.76 & 0.67 & 0.82 & 0.76 & 0.82 & \\
n = 2500\\
Bias & 0.008 & 0.601 & 0.744 & 0.006 & 0.007 & 0.744 & 0.006 & 0.592\\
Variance & 0.030 & 0.025 & 0.047 & 0.020 & 0.040 & 0.038 & 0.030 & 0.009\\
MSE & 0.030 & 0.386 & 0.600 & 0.020 & 0.040 & 0.591 & 0.030 & 0.359\\
avg c-stat & 0.67 & 0.51 & 0.76 & 0.67 & 0.81 & 0.76 & 0.81 & \\
\hline
\end{tabular}
\end{table}
```
```{=tex}
\begin{table}
\caption{Table 2. Simulation experiment 1, with results based on an analysis using
subclassification in which strata are defined by quintiles of the estimated propensity
score.}
\begin{tabular}{l r r r r r r r r}
\hline
\multicolumn{8}{c}{Variable(s) in propensity score model}\\
\hline
& $X_{1}$ & $X_{2}$ & $X_{3}$ & $X_{1}$, $X_{2}$ & $X_{1}$, $X_{3}$ & $X_{2}$, $X_{3}$ & $X_{1}$, $X_{2}$, $X_{3}$ & None\\
\hline
n = 500\\
Bias & 0.027 & 0.607 & 0.802 & 0.030 & 0.037 & 0.801 & 0.033 & 0.599\\
Variance & 0.022 & 0.018 & 0.050 & 0.016 & 0.048 & 0.039 & 0.038 & 0.040\\
MSE & 0.023 & 0.386 & 0.693 & 0.016 & 0.050 & 0.680 & 0.039 & 0.398\\
n = 2500\\
Bias & 0.029 & 0.597 & 0.763 & 0.029 & 0.063 & 0.762 & 0.061 & 0.592\\
Variance & 0.004 & 0.003 & 0.011 & 0.003 & 0.012 & 0.009 & 0.010 & 0.009\\
MSE & 0.005 & 0.360 & 0.594 & 0.004 & 0.016 & 0.589 & 0.014 & 0.359\\
\hline
\end{tabular}
\end{table}
```
```{=tex}
\begin{table}
\caption{Table 3. Sensitivity analysis of simulation study 1. We consider nine different perturbations of the simulation parameters. Results are from 1,000 simulations of
data (n = 500), using a parametric spline to adjust for the estimated propensity score.}
\begin{tabular}{l r r r r r r r r}
\hline
\multicolumn{8}{c}{Variable(s) in propensity score model}\\
\hline
& $X_{1}$ & $X_{2}$ & $X_{3}$ & $X_{1}$, $X_{2}$ & $X_{1}$, $X_{3}$ & $X_{2}$, $X_{3}$ & $X_{1}$, $X_{2}$, $X_{3}$ & None\\
\hline
std1 = 0.5\\
Bias & -0.003 & 0.281 & 0.364 & -0.006 & -0.001 & 0.360 & -0.005 & 0.278\\
Variance & 0.031 & 0.021 & 0.047 & 0.021 & 0.041 & 0.037 & 0.029 & 0.037\\
MSE & 0.031 & 0.101 & 0.180 & 0.021 & 0.041 & 0.166 & 0.029 & 0.115\\
std1 = 1.5\\
Bias & 0.008 & 0.855 & 1.009 & 0.007 & 0.004 & 1.013 & 0.008 & 0.847\\
Variance & 0.033 & 0.028 & 0.048 & 0.024 & 0.046 & 0.040 & 0.035 & 0.041\\
MSE & 0.033 & 0.759 & 1.066 & 0.024 & 0.046 & 1.066 & 0.035 & 0.759\\
std2 = 0.5\\
Bias & 0.003 & 0.595 & 0.738 & -0.001 & 0.001 & 0.734 & -0.004 & 0.599\\
Variance & 0.007 & 0.015 & 0.020 & 0.005 & 0.011 & 0.018 & 0.009 & 0.017\\
MSE & 0.007 & 0.369 & 0.564 & 0.005 & 0.011 & 0.557 & 0.009 & 0.376\\
std2 = 1.5\\
Bias & 0.015 & 0.602 & 0.751 & 0.014 & 0.019 & 0.748 & 0.014 & 0.586\\
Variance & 0.106 & 0.055 & 0.146 & 0.090 & 0.147 & 0.122 & 0.121 & 0.137\\
MSE & 0.106 & 0.418 & 0.710 & 0.090 & 0.147 & 0.680 & 0.121 & 0.480\\
std3 = 0.5\\
Bias & -0.003 & 0.698 & 0.732 & -0.004 & -0.002 & 0.732 & -0.003 & 0.691\\
Variance & 0.032 & 0.025 & 0.045 & 0.022 & 0.038 & 0.035 & 0.027 & 0.041\\
MSE & 0.032 & 0.512 & 0.581 & 0.022 & 0.038 & 0.571 & 0.027 & 0.518\\
std3 = 1.5\\
Bias & -0.001 & 0.495 & 0.734 & -0.004 & -0.006 & 0.734 & -0.008 & 5.07\\
Variance & 0.027 & 0.022 & 0.055 & 0.021 & 0.049 & 0.044 & 0.037 & 0.041\\
MSE & 0.028 & 0.267 & 0.593 & 0.021 & 0.049 & 0.582 & 0.038 & 0.518\\
a4 = 0.25\\
Bias & -0.004 & 0.593 & 0.731 & 0.000 & -0.007 & 0.737 & 0.000 & 0.600\\
Variance & 0.032 & 0.025 & 0.049 & 0.023 & 0.044 & 0.039 & 0.033 & 0.040\\
MSE & 0.032 & 0.377 & 0.583 & 0.023 & 0.044 & 0.582 & 0.033 & 0.399\\
a4 = 1\\
Bias & 0.004 & 0.601 & 0.735 & 0.005 & 0.004 & 0.735 & 0.005 & 0.597\\
Variance & 0.033 & 0.027 & 0.052 & 0.023 & 0.047 & 0.042 & 0.035 & 0.040\\
MSE & 0.033 & 0.387 & 0.592 & 0.023 & 0.047 & 0.582 & 0.035 & 0.397\\
b0 = 1\\
Bias & -0.012 & 0.562 & 0.687 & -0.010 & -0.015 & 0.690 & -0.012 & 0.575\\
Variance & 0.034 & 0.021 & 0.052 & 0.022 & 0.047 & 0.039 & 0.032 & 0.041\\
MSE & 0.035 & 0.337 & 0.525 & 0.022 & 0.047 & 0.515 & 0.032 & 0.372\\
\hline
\end{tabular}
\end{table}
```
\subsection{Experiment 2}
We were not able to replicate the figures of experiment 2.
\FloatBarrier
\section{Discussion}
\subsection{Replicability}
In general, the data-generating mechanism is easy to replicate. It is clearly described in the original paper and it can be easily coded. Also, experiment 1 can be replicated with only a few replication degrees of freedom. However, availability of open source code will improve its replicability. The initial lack of explanation for the c-statistic or the issue with the model with splines could have been solved easily if the code was available.
Experiment two is more difficult to replicate, or, at least, the figures on the paper. The model estimation and the computation of the measures are similar to experiment one. However, once these results are obtained, it is necessary to use some transformations that are not explained at all in the paper. Also, as stated previously, replicating a complex figure is not an easy task.In fact, we were not able to do so. The fact that we were not able to replicate experiment 2 does not mean that it is not possible. It is likely that more experienced researchers in the field would be able to replicate experiment 2.
\subsection{Replicator degrees of freedom}
In experiment 1, we needed to guess some concepts due to the lack of information on the article. However, the choices were straightforward and it we were able to replicate the experiment.
In experiment 2, the lack of information about some concepts and scales did not allow us to replicate the experiment. Probably, when writing the paper the authors addressed an audience with a higher experience in the field than the authors of this report. However, it is always a good practice to write a paper as self-explanatory as possible. This will help reaching a wider audience and reduce the replicator degrees of freedom. At the same time, these technical details are not relevant for understanding the main concept or the message the authors want to present. They are mainly necessary for the purpose of replication.
\subsection{Equivalence of results}
The equivalence of results in experiment 1 is extremely good. Despite using a different R version and code, we were able to obtain very similar same tables.
We cannot judge the results of experiment 2 because we were not able to obtain the same figures.
\section{Comparison with other pair performing the same replication study}
The other pair was able to perform an even more successful replication. Both groups were able to replicate experiment 1 with similar issues that we solved similarly yet independently. On the other hand, the other pair was able to replicate figure 2 and figure 3 from the original paper and almost able to replicate figure 4.
A discussion between both groups leaded to the following conclusions:
The order in which the method, measures and results are presented in the paper can be improved in order to facilitate the understanding and replicability of the research.
The original paper omit some basic information assuming it is common knowledge. However, we do not consider it common knowledge. This fact could explain why one of the groups was able to replicate more than the other. Actually, in latter versions of the paper they included explanations of some concepts (e.g. meaning of c-statistic).
The first experiment is explained with more detail, which makes it easier to replicate. On the other hand, both groups consider that experiment two contains some concepts that are not fully explained, which hurts its replicability.
Both groups consider that the original code of the research could be available publicly since it is coded in an open source programming laguage (R), and it does not require new functions that could be under copyrigth. Publishing the code will improve the replicability and transparency of the research.
\section{Contributions}
Authors made the following contributions according to the CRediT
framework <https://casrai.org/credit/>
Primary Replicator:
- Data Curation\
- Formal Analysis (supporting)\
- Investigation\
- Software\
- Visualization (lead)\
- Writing - Original Draft Preparation\
- Writing - Review & Editing
Co-Pilot:
- Formal Analysis (lead)\
- Investigation\
- Software (supporting)\
- Visualization (supporting)\
- Validation\
- Writing - Review & Editing
\newpage
```{=tex}
\section*{References}
\begingroup
\hphantom{x}
\setlength{\parindent}{-0.5in}
\setlength{\leftskip}{0.5in}
```
::: {#refs custom-style="Bibliography"}
:::
```{=tex}
\FloatBarrier
\endgroup
\newpage
```
\section\*{Appendix}
\subsection{Code organization}
The code and the files associated are organized in the form of a
research compendium which can be found in the following git repository
`https://github.com/replisims/Brookhart-2006-FJ`
```{r code_structure, cache = FALSE}
# which R packages and versions?
if ("fs" %in% installed.packages()) fs::dir_tree(path = ".", recurse = TRUE)
```
\subsubsection\*{Reproducibility Information}
This report was last updated on `r Sys.time()`. The simulation
replication was conducted using the following computational environment
and dependencies:
\FloatBarrier
```{r colophon, cache = FALSE}
# which R packages and versions?
if ("devtools" %in% installed.packages()) devtools::session_info()
```
The current Git commit details are:
```{r git}
# what commit is this file at?
if ("git2r" %in% installed.packages() & git2r::in_repository(path = ".")) git2r::repository(here::here())
```