forked from opain/GenoPred
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Z-test_vs_ANOVA_V3.html
780 lines (650 loc) · 25.4 KB
/
Z-test_vs_ANOVA_V3.html
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
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8" />
<meta name="generator" content="pandoc" />
<meta http-equiv="X-UA-Compatible" content="IE=EDGE" />
<title>Estimating significance of prediction differences</title>
<script src="site_libs/jquery-1.11.3/jquery.min.js"></script>
<meta name="viewport" content="width=device-width, initial-scale=1" />
<link href="site_libs/bootstrap-3.3.5/css/united.min.css" rel="stylesheet" />
<script src="site_libs/bootstrap-3.3.5/js/bootstrap.min.js"></script>
<script src="site_libs/bootstrap-3.3.5/shim/html5shiv.min.js"></script>
<script src="site_libs/bootstrap-3.3.5/shim/respond.min.js"></script>
<style>h1 {font-size: 34px;}
h1.title {font-size: 38px;}
h2 {font-size: 30px;}
h3 {font-size: 24px;}
h4 {font-size: 18px;}
h5 {font-size: 16px;}
h6 {font-size: 12px;}
code {color: inherit; background-color: rgba(0, 0, 0, 0.04);}
pre:not([class]) { background-color: white }</style>
<script src="site_libs/jqueryui-1.11.4/jquery-ui.min.js"></script>
<link href="site_libs/tocify-1.9.1/jquery.tocify.css" rel="stylesheet" />
<script src="site_libs/tocify-1.9.1/jquery.tocify.js"></script>
<script src="site_libs/navigation-1.1/tabsets.js"></script>
<link href="site_libs/highlightjs-9.12.0/default.css" rel="stylesheet" />
<script src="site_libs/highlightjs-9.12.0/highlight.js"></script>
<link href="site_libs/font-awesome-5.1.0/css/all.css" rel="stylesheet" />
<link href="site_libs/font-awesome-5.1.0/css/v4-shims.css" rel="stylesheet" />
<style type="text/css">
code{white-space: pre-wrap;}
span.smallcaps{font-variant: small-caps;}
span.underline{text-decoration: underline;}
div.column{display: inline-block; vertical-align: top; width: 50%;}
div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
ul.task-list{list-style: none;}
</style>
<style type="text/css">code{white-space: pre;}</style>
<script type="text/javascript">
if (window.hljs) {
hljs.configure({languages: []});
hljs.initHighlightingOnLoad();
if (document.readyState && document.readyState === "complete") {
window.setTimeout(function() { hljs.initHighlighting(); }, 0);
}
}
</script>
<style type = "text/css">
.main-container {
max-width: 940px;
margin-left: auto;
margin-right: auto;
}
img {
max-width:100%;
}
.tabbed-pane {
padding-top: 12px;
}
.html-widget {
margin-bottom: 20px;
}
button.code-folding-btn:focus {
outline: none;
}
summary {
display: list-item;
}
pre code {
padding: 0;
}
</style>
<style type="text/css">
.dropdown-submenu {
position: relative;
}
.dropdown-submenu>.dropdown-menu {
top: 0;
left: 100%;
margin-top: -6px;
margin-left: -1px;
border-radius: 0 6px 6px 6px;
}
.dropdown-submenu:hover>.dropdown-menu {
display: block;
}
.dropdown-submenu>a:after {
display: block;
content: " ";
float: right;
width: 0;
height: 0;
border-color: transparent;
border-style: solid;
border-width: 5px 0 5px 5px;
border-left-color: #cccccc;
margin-top: 5px;
margin-right: -10px;
}
.dropdown-submenu:hover>a:after {
border-left-color: #adb5bd;
}
.dropdown-submenu.pull-left {
float: none;
}
.dropdown-submenu.pull-left>.dropdown-menu {
left: -100%;
margin-left: 10px;
border-radius: 6px 0 6px 6px;
}
</style>
<script type="text/javascript">
// manage active state of menu based on current page
$(document).ready(function () {
// active menu anchor
href = window.location.pathname
href = href.substr(href.lastIndexOf('/') + 1)
if (href === "")
href = "index.html";
var menuAnchor = $('a[href="' + href + '"]');
// mark it active
menuAnchor.tab('show');
// if it's got a parent navbar menu mark it active as well
menuAnchor.closest('li.dropdown').addClass('active');
// Navbar adjustments
var navHeight = $(".navbar").first().height() + 15;
var style = document.createElement('style');
var pt = "padding-top: " + navHeight + "px; ";
var mt = "margin-top: -" + navHeight + "px; ";
var css = "";
// offset scroll position for anchor links (for fixed navbar)
for (var i = 1; i <= 6; i++) {
css += ".section h" + i + "{ " + pt + mt + "}\n";
}
style.innerHTML = "body {" + pt + "padding-bottom: 40px; }\n" + css;
document.head.appendChild(style);
});
</script>
<!-- tabsets -->
<style type="text/css">
.tabset-dropdown > .nav-tabs {
display: inline-table;
max-height: 500px;
min-height: 44px;
overflow-y: auto;
border: 1px solid #ddd;
border-radius: 4px;
}
.tabset-dropdown > .nav-tabs > li.active:before {
content: "";
font-family: 'Glyphicons Halflings';
display: inline-block;
padding: 10px;
border-right: 1px solid #ddd;
}
.tabset-dropdown > .nav-tabs.nav-tabs-open > li.active:before {
content: "";
border: none;
}
.tabset-dropdown > .nav-tabs.nav-tabs-open:before {
content: "";
font-family: 'Glyphicons Halflings';
display: inline-block;
padding: 10px;
border-right: 1px solid #ddd;
}
.tabset-dropdown > .nav-tabs > li.active {
display: block;
}
.tabset-dropdown > .nav-tabs > li > a,
.tabset-dropdown > .nav-tabs > li > a:focus,
.tabset-dropdown > .nav-tabs > li > a:hover {
border: none;
display: inline-block;
border-radius: 4px;
background-color: transparent;
}
.tabset-dropdown > .nav-tabs.nav-tabs-open > li {
display: block;
float: none;
}
.tabset-dropdown > .nav-tabs > li {
display: none;
}
</style>
<!-- code folding -->
<style type="text/css">
#TOC {
margin: 25px 0px 20px 0px;
}
@media (max-width: 768px) {
#TOC {
position: relative;
width: 100%;
}
}
@media print {
.toc-content {
/* see https://github.com/w3c/csswg-drafts/issues/4434 */
float: right;
}
}
.toc-content {
padding-left: 30px;
padding-right: 40px;
}
div.main-container {
max-width: 1200px;
}
div.tocify {
width: 20%;
max-width: 260px;
max-height: 85%;
}
@media (min-width: 768px) and (max-width: 991px) {
div.tocify {
width: 25%;
}
}
@media (max-width: 767px) {
div.tocify {
width: 100%;
max-width: none;
}
}
.tocify ul, .tocify li {
line-height: 20px;
}
.tocify-subheader .tocify-item {
font-size: 0.90em;
}
.tocify .list-group-item {
border-radius: 0px;
}
</style>
</head>
<body>
<div class="container-fluid main-container">
<!-- setup 3col/9col grid for toc_float and main content -->
<div class="row">
<div class="col-sm-12 col-md-4 col-lg-3">
<div id="TOC" class="tocify">
</div>
</div>
<div class="toc-content col-sm-12 col-md-8 col-lg-9">
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<div class="navbar-header">
<button type="button" class="navbar-toggle collapsed" data-toggle="collapse" data-target="#navbar">
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="index.html">GenoPred</a>
</div>
<div id="navbar" class="navbar-collapse collapse">
<ul class="nav navbar-nav">
</ul>
<ul class="nav navbar-nav navbar-right">
<li>
<a href="https://github.com/opain/GenoPred">
<span class="fas fa-github fa-lg"></span>
</a>
</li>
</ul>
</div><!--/.nav-collapse -->
</div><!--/.container -->
</div><!--/.navbar -->
<div id="header">
<h1 class="title toc-ignore">Estimating significance of prediction differences</h1>
</div>
<hr />
<p>(under development)</p>
<p>This analysis was to compare different methods for estimating the statistical significance of prediction differences across polygenic scoring methods, and models in general.</p>
<p>Here I simulate data with one continuous outcome, and two correlated continuous sets of predictions The correlation between the predictors and the outcome is varied, and the correlation between the predictors themselves is varied.</p>
<p>As I do in my other analyses, I estimate the correlation between the predictors and the outcome to determine their predictive utility. Then I compare different methods by comparing the observed-predicted correlations. Several method comparing correlation are considered:</p>
<ul>
<li>Two-sample Z-test
<ul>
<li>Compares estimates from two populations and does not account for the correlation between predictors</li>
</ul></li>
<li>Permutation based
<ul>
<li>Randomises the phenotype, retaining the correlation between predictors, and then estimates the number times the difference in correlation is larger than the observed difference.</li>
</ul></li>
<li>Cox test
<ul>
<li>Method for comparing non-nested models</li>
</ul></li>
<li>Pearson and Filon’s (1898):
<ul>
<li>Method for comparing correlations between variables that are correlated and measured in a single sample</li>
<li>Implemented using cocor.dep.groups.overlap function in the cocor package</li>
<li>Other methods implemented by this function are highly concordant</li>
</ul></li>
<li>Fisher r-to-z :
<ul>
<li>Method for comparing correlations between variables that are independent and measures in different samples.</li>
<li>Implemented using psych package</li>
<li>Accounts for non-normal error of correlations</li>
</ul></li>
<li>Williams test :
<ul>
<li>Method for comparing correlations between variables that are dependent and measured in the same sample.</li>
<li>Implemented using psych package</li>
<li>Accounts for non-normal error of correlations</li>
</ul></li>
</ul>
<p>We will simulate the following scenarios:</p>
<ul>
<li>Estimates from two independent samples</li>
<li>Estimates from one sample, but predictors are uncorrelated</li>
<li>Estimates from one sample, and predictors are highly correlated</li>
</ul>
<hr />
<div id="estimates-from-two-independent-samples" class="section level1">
<h1><span class="header-section-number">1</span> Estimates from two independent samples</h1>
<pre class="r"><code>library(data.table)
set.seed(1)
# Sample 1
N<-200
y1<-rnorm(N)
x1<-scale(y1+rnorm(N,0,3))
dat1<-data.table(y1,x1)
cor(dat1)</code></pre>
<pre><code>## y1 V1
## y1 1.0000000 0.2743249
## V1 0.2743249 1.0000000</code></pre>
<pre class="r"><code># Sample 2
N<-200
y2<-rnorm(N)
x2<-scale(y2+rnorm(N,0,10))
dat2<-data.table(y2,x2)
cor(dat2)</code></pre>
<pre><code>## y2 V1
## y2 1.00000000 0.06274804
## V1 0.06274804 1.00000000</code></pre>
<pre class="r"><code>### Derive models using each predictor
mod1<-lm(y1 ~ x1, data=dat1)
mod2<-lm(y2 ~ x2, data=dat2)
dat1$pred1<-predict(mod1, dat1)
dat2$pred2<-predict(mod2, dat2)
dat_cor<-data.frame(model1='x1',
model2='x2',
cor_x1_x2=NA,
cor_y_x1=coef(summary(mod1))[2,1],
cor_y_x1_se=coef(summary(mod1))[2,2],
cor_y_x2=coef(summary(mod2))[2,1],
cor_y_x2_se=coef(summary(mod2))[2,2],
cor_diff=coef(summary(mod1))[2,1]-coef(summary(mod2))[2,1])
### Test difference between predictors using a Z-test
dat_cor$cor_diff_Ztest_P<-pnorm(-(dat_cor$cor_diff/sqrt((dat_cor$cor_y_x1_se^2)+(dat_cor$cor_y_x2_se^2))))
### Test difference between predictors using a permutation test
n_perm<-500
diff<-NULL
for(i in 1:n_perm){
y1_sample<-sample(y1)
y2_sample<-sample(y2)
mod1_i<-lm(y1_sample ~ x1, data=dat1)
mod2_i<-lm(y2_sample ~ x2, data=dat2)
diff_i<-coef(summary(mod1_i))[2,1]-coef(summary(mod2_i))[2,1]
diff<-c(diff,diff_i)
}
dat_cor$cor_diff_perm_P<-sum(diff > dat_cor$cor_diff[1])/n_perm
### Test difference between predictors using coxtest
# Not possible as different samples
### Test difference between predictors using Fisher's Z transformation
library(cocor)
dat_cor$cocor_fisherZ_P<-cocor.indep.groups(r1.jk=dat_cor$cor_y_x1, r2.hm=dat_cor$cor_y_x2, n1=N, n2=N, alternative='greater')@fisher1925$p.value
library(psych)
dat_cor$psych_fisherZ_P<-paired.r(xy=dat_cor$cor_y_x1, xz=dat_cor$cor_y_x2, n=N, n2=N, twotailed=F)$p
dat_cor</code></pre>
<pre><code>## model1 model2 cor_x1_x2 cor_y_x1 cor_y_x1_se cor_y_x2 cor_y_x2_se
## 1 x1 x2 NA 0.2548745 0.06349504 0.06714255 0.07589417
## cor_diff cor_diff_Ztest_P cor_diff_perm_P cocor_fisherZ_P psych_fisherZ_P
## 1 0.1877319 0.02890093 0.046 0.02747978 0.02747978</code></pre>
<p>The Fisher’s r-to-z transformation is a commonly used approach for comparing correlations from different samples. The two-sample z-test does not account for the non-normal error distribution of Pearson correlations. Results indicate the fisher r-to-z method is concordant with the two-sample z-test. The permutation-based approach is more conservative than other methods.</p>
<hr />
</div>
<div id="estimates-from-one-sample-but-predictors-are-uncorrelated" class="section level1">
<h1><span class="header-section-number">2</span> Estimates from one sample, but predictors are uncorrelated</h1>
<pre class="r"><code>library(data.table)
set.seed(1)
N<-200
y<-rnorm(N)
x1<-as.numeric(scale(y+rnorm(N,0,3)))
x2<-as.numeric(scale(y+rnorm(N,0,10)))
dat<-data.table(y,x1,x2)
cor(dat)</code></pre>
<pre><code>## y x1 x2
## y 1.0000000 0.27432489 0.15351734
## x1 0.2743249 1.00000000 0.01848117
## x2 0.1535173 0.01848117 1.00000000</code></pre>
<pre class="r"><code>### Derive models using each predictor
mod1<-lm(y ~ x1, data=dat)
mod2<-lm(y ~ x2, data=dat)
dat$pred1<-predict(mod1, dat)
dat$pred2<-predict(mod2, dat)
dat_cor<-data.frame(model1='x1',
model2='x2',
cor_x1_x2=cor(dat$x1,dat$x2),
cor_y_x1=coef(summary(mod1))[2,1],
cor_y_x1_se=coef(summary(mod1))[2,2],
cor_y_x2=coef(summary(mod2))[2,1],
cor_y_x2_se=coef(summary(mod2))[2,2],
cor_diff=coef(summary(mod1))[2,1]-coef(summary(mod2))[2,1])
### Test difference between predictors using a Z-test
dat_cor$cor_diff_Ztest_P<-pnorm(-(dat_cor$cor_diff/sqrt((dat_cor$cor_y_x1_se^2)+(dat_cor$cor_y_x2_se^2))))
### Test difference between predictors using a permutation test
n_perm<-500
diff<-NULL
for(i in 1:n_perm){
y_sample<-sample(y)
mod1_i<-lm(y_sample ~ x1, data=dat)
mod2_i<-lm(y_sample ~ x2, data=dat)
diff_i<-coef(summary(mod1_i))[2,1]-coef(summary(mod2_i))[2,1]
diff<-c(diff,diff_i)
}
dat_cor$cor_diff_perm_P<-sum(diff > dat_cor$cor_diff[1])/n_perm
### Test difference between predictors using coxtest
library(lmtest)
dat_cor$cox_diff_P<-coxtest(mod1, mod2)$P[2]
### Test difference between predictors using Pearson
library(cocor)
dat_cor$pearson_diff_P<-cocor.dep.groups.overlap(dat_cor$cor_y_x1, dat_cor$cor_y_x2, dat_cor$cor_x1_x2, N,alternative='greater')@pearson1898$p.value
### Test difference between predictors using Williams's Test
library(psych)
dat_cor$williams_diff_P<-paired.r(xy=dat_cor$cor_y_x1, xz=dat_cor$cor_y_x2, yz=dat_cor$cor_x1_x2, n=N, twotailed=F)$p
dat_cor</code></pre>
<pre><code>## model1 model2 cor_x1_x2 cor_y_x1 cor_y_x1_se cor_y_x2 cor_y_x2_se cor_diff
## 1 x1 x2 0.01848117 0.2548745 0.06349504 0.1426325 0.06524537 0.112242
## cor_diff_Ztest_P cor_diff_perm_P cox_diff_P pearson_diff_P williams_diff_P
## 1 0.1088132 0.112 0 0.1205402 0.1230892</code></pre>
<p>Apart fromt the coxtest method, results are similar across methods, with pearson and williams methods being highly concordant.</p>
<hr />
</div>
<div id="estimates-from-one-sample-and-predictors-are-highly-correlated" class="section level1">
<h1><span class="header-section-number">3</span> Estimates from one sample, and predictors are highly correlated</h1>
<pre class="r"><code>library(data.table)
set.seed(1)
N<-200
y<-rnorm(N)
set.seed(2)
x1<-as.numeric(scale(y+rnorm(N,0,3)))
set.seed(3)
x2<-as.numeric(scale(x1+rnorm(N,0,1)))
dat<-data.table(y,x1,x2)
cor(dat)</code></pre>
<pre><code>## y x1 x2
## y 1.0000000 0.2813946 0.1928833
## x1 0.2813946 1.0000000 0.6928299
## x2 0.1928833 0.6928299 1.0000000</code></pre>
<pre class="r"><code>### Derive models using each predictor
mod1<-lm(y ~ x1, data=dat)
mod2<-lm(y ~ x2, data=dat)
dat$pred1<-predict(mod1, dat)
dat$pred2<-predict(mod2, dat)
dat_cor<-data.frame(model1='x1',
model2='x2',
cor_x1_x2=cor(dat$x1,dat$x2),
cor_y_x1=coef(summary(mod1))[2,1],
cor_y_x1_se=coef(summary(mod1))[2,2],
cor_y_x2=coef(summary(mod2))[2,1],
cor_y_x2_se=coef(summary(mod2))[2,2],
cor_diff=coef(summary(mod1))[2,1]-coef(summary(mod2))[2,1])
### Test difference between predictors using a Z-test
dat_cor$cor_diff_Ztest_P<-pnorm(-(dat_cor$cor_diff/sqrt((dat_cor$cor_y_x1_se^2)+(dat_cor$cor_y_x2_se^2))))
### Test difference between predictors using a permutation test
n_perm<-500
diff<-NULL
for(i in 1:n_perm){
y_sample<-sample(y)
mod1_i<-lm(y_sample ~ x1, data=dat)
mod2_i<-lm(y_sample ~ x2, data=dat)
diff_i<-coef(summary(mod1_i))[2,1]-coef(summary(mod2_i))[2,1]
diff<-c(diff,diff_i)
}
dat_cor$cor_diff_perm_P<-sum(diff > dat_cor$cor_diff[1])/n_perm
### Test difference between predictors using coxtest
library(lmtest)
dat_cor$cox_diff_P<-coxtest(mod1, mod2)$P[2]
### Test difference between predictors using Pearson
library(cocor)
dat_cor$pearson_diff_P<-cocor.dep.groups.overlap(dat_cor$cor_y_x1, dat_cor$cor_y_x2, dat_cor$cor_x1_x2, N,alternative='greater')@pearson1898$p.value
### Test difference between predictors using Williams's Test
library(psych)
dat_cor$williams_diff_P<-paired.r(xy=dat_cor$cor_y_x1, xz=dat_cor$cor_y_x2, yz=dat_cor$cor_x1_x2, n=N, twotailed=F)$p
### Test difference between predictors using ROC test
library(pROC)</code></pre>
<pre><code>## Type 'citation("pROC")' for a citation.</code></pre>
<pre><code>##
## Attaching package: 'pROC'</code></pre>
<pre><code>## The following objects are masked from 'package:stats':
##
## cov, smooth, var</code></pre>
<pre class="r"><code>mod1_roc<-roc(y ~ x1)</code></pre>
<pre><code>## Warning in roc.default(response, predictors[, 1], ...): 'response' has more
## than two levels. Consider setting 'levels' explicitly or using 'multiclass.roc'
## instead</code></pre>
<pre><code>## Setting levels: control = -2.2146998871775, case = -1.98935169586337</code></pre>
<pre><code>## Setting direction: controls < cases</code></pre>
<pre class="r"><code>mod2_roc<-roc(y ~ x2)</code></pre>
<pre><code>## Warning in roc.default(response, predictors[, 1], ...): 'response' has more
## than two levels. Consider setting 'levels' explicitly or using 'multiclass.roc'
## instead</code></pre>
<pre><code>## Setting levels: control = -2.2146998871775, case = -1.98935169586337
## Setting direction: controls < cases</code></pre>
<pre class="r"><code>dat_cor$mod1_auc<-mod1_roc$auc
dat_cor$mod2_auc<-mod2_roc$auc
dat_cor$roc_diff_P<-roc.test(mod1_roc,mod2_roc, paired=T, alternative='greater', method='bootstrap')$p.value</code></pre>
<pre><code>## Warning in roc.test.roc(mod1_roc, mod2_roc, paired = T, alternative =
## "greater", : roc.test() of two ROC curves with AUC == 1 has always p.value = 1
## and can be misleading.</code></pre>
<pre class="r"><code>dat_cor</code></pre>
<pre><code>## model1 model2 cor_x1_x2 cor_y_x1 cor_y_x1_se cor_y_x2 cor_y_x2_se
## 1 x1 x2 0.6928299 0.2614429 0.06336002 0.1792073 0.06478817
## cor_diff cor_diff_Ztest_P cor_diff_perm_P cox_diff_P pearson_diff_P
## 1 0.08223559 0.1820774 0.068 2.185878e-06 0.06303985
## williams_diff_P mod1_auc mod2_auc roc_diff_P
## 1 0.06448287 1 1 0.5</code></pre>
<p>Again the coxtest method is very different from other methods. The two-sample Z test is now deviating from the other methods because it doesn’t account for the correlation between predictors. The results for the permutation, pearson and williams method are highly concordant.</p>
<p>The psych package is a solid package and the Williams method is recommended by Steiger who worked alot in this area. It is also faster than the permutation based approach. From here on use the Williams test to test for significant differences between correlations between outcomes and correlated predictors.</p>
<hr />
</div>
<div id="under-development" class="section level1">
<h1><span class="header-section-number">4</span> Under development…</h1>
<p>Another method that is used to compare models is to compare AUC curves. Ths can be implemented using the pROC package in R. This test is only suitable for binary outcomes. For comparison, compare the AUC method to the permutation and wiliams methods.</p>
<hr />
<div id="estimates-from-one-sample-and-predictors-are-highly-correlated-1" class="section level2">
<h2><span class="header-section-number">4.1</span> Estimates from one sample, and predictors are highly correlated</h2>
<pre class="r"><code>library(data.table)
set.seed(1)
N<-200
y<-rbinom(n=N, size=1, prob=0.5)
set.seed(2)
x1<-as.numeric(scale(y+rnorm(N,0,2)))
set.seed(3)
x2<-as.numeric(scale(x1+rnorm(N,0,1)))
dat<-data.table(y,x1,x2)
cor(dat)</code></pre>
<pre><code>## y x1 x2
## y 1.0000000 0.1747443 0.1361085
## x1 0.1747443 1.0000000 0.6951187
## x2 0.1361085 0.6951187 1.0000000</code></pre>
<pre class="r"><code>### Derive models using each predictor
mod1<-glm(y ~ x1, data=dat, family='binomial')
mod2<-glm(y ~ x2, data=dat, family='binomial')
dat$pred1<-predict(mod1, dat)
dat$pred2<-predict(mod2, dat)
dat_cor<-data.frame(model1='x1',
model2='x2',
cor_x1_x2=cor(dat$x1,dat$x2),
cor_y_x1=coef(summary(mod1))[2,1],
cor_y_x1_se=coef(summary(mod1))[2,2],
cor_y_x2=coef(summary(mod2))[2,1],
cor_y_x2_se=coef(summary(mod2))[2,2],
cor_diff=coef(summary(mod1))[2,1]-coef(summary(mod2))[2,1])
### Test difference between predictors using a permutation test
n_perm<-500
diff<-NULL
for(i in 1:n_perm){
y_sample<-sample(y)
mod1_i<-glm(y_sample ~ x1, data=dat, family='binomial')
mod2_i<-glm(y_sample ~ x2, data=dat, family='binomial')
diff_i<-coef(summary(mod1_i))[2,1]-coef(summary(mod2_i))[2,1]
diff<-c(diff,diff_i)
}
dat_cor$cor_diff_perm_P<-sum(diff > dat_cor$cor_diff[1])/n_perm
### Test difference between predictors using Williams's Test
library(psych)
dat_cor$williams_diff_P<-paired.r(xy=dat_cor$cor_y_x1, xz=dat_cor$cor_y_x2, yz=dat_cor$cor_x1_x2, n=N, twotailed=F)$p
### Test difference between predictors using ROC test
library(pROC)
mod1_roc<-roc(y ~ x1)</code></pre>
<pre><code>## Setting levels: control = 0, case = 1</code></pre>
<pre><code>## Setting direction: controls < cases</code></pre>
<pre class="r"><code>mod2_roc<-roc(y ~ x2)</code></pre>
<pre><code>## Setting levels: control = 0, case = 1
## Setting direction: controls < cases</code></pre>
<pre class="r"><code>dat_cor$mod1_auc<-mod1_roc$auc
dat_cor$mod2_auc<-mod2_roc$auc
dat_cor$roc_diff_P<-roc.test(mod1_roc,mod2_roc, paired=T, alternative='greater')$p.value
dat_cor</code></pre>
<pre><code>## model1 model2 cor_x1_x2 cor_y_x1 cor_y_x1_se cor_y_x2 cor_y_x2_se cor_diff
## 1 x1 x2 0.6951187 0.359497 0.1473043 0.2777717 0.1455746 0.0817253
## cor_diff_perm_P williams_diff_P mod1_auc mod2_auc roc_diff_P
## 1 0.216 0.0586538 0.5869348 0.5847339 0.4737637</code></pre>
<p>This analysis raises some concerns about the validity of the William’s test when the outcome is binary. We should check the false positive rate of the Williams test by testing how many <0.05 tests there are under the null.</p>
<hr />
</div>
</div>
</div>
</div>
</div>
<script>
// add bootstrap table styles to pandoc tables
function bootstrapStylePandocTables() {
$('tr.odd').parent('tbody').parent('table').addClass('table table-condensed');
}
$(document).ready(function () {
bootstrapStylePandocTables();
});
</script>
<!-- tabsets -->
<script>
$(document).ready(function () {
window.buildTabsets("TOC");
});
$(document).ready(function () {
$('.tabset-dropdown > .nav-tabs > li').click(function () {
$(this).parent().toggleClass('nav-tabs-open');
});
});
</script>
<!-- code folding -->
<script>
$(document).ready(function () {
// move toc-ignore selectors from section div to header
$('div.section.toc-ignore')
.removeClass('toc-ignore')
.children('h1,h2,h3,h4,h5').addClass('toc-ignore');
// establish options
var options = {
selectors: "h1,h2,h3",
theme: "bootstrap3",
context: '.toc-content',
hashGenerator: function (text) {
return text.replace(/[.\\/?&!#<>]/g, '').replace(/\s/g, '_');
},
ignoreSelector: ".toc-ignore",
scrollTo: 0
};
options.showAndHide = true;
options.smoothScroll = true;
// tocify
var toc = $("#TOC").tocify(options).data("toc-tocify");
});
</script>
<!-- dynamically load mathjax for compatibility with self-contained -->
<script>
(function () {
var script = document.createElement("script");
script.type = "text/javascript";
script.src = "https://mathjax.rstudio.com/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML";
document.getElementsByTagName("head")[0].appendChild(script);
})();
</script>
</body>
</html>