Bayesian Analysis

Module II: Leaves and Trees

Dr. Mark Williamson
DaCCoTA
University of North Dakota
 Introduction
• Last time, we covered the history and • Today we’ll cover more details about the
basic theory of Bayesian Analysis, as computational process
well as comparison to Frequentist • We will also look at examples in R, SAS, and
methods SPSS for t-tests, ANOVA, and regressions
 Theory
Bayes’ Theorem
• Updates probabilities (degrees of belief) after obtaining new data
Bayes Theorem:
𝑃 𝑑𝑎𝑡𝑎 ℎ𝑦𝑝 × 𝑃(ℎ𝑦𝑝)
𝑃 ℎ𝑦𝑝 𝑑𝑎𝑡𝑎 =
𝑃(𝑑𝑎𝑡𝑎)
Bayes Theorem:
𝑃 𝐵 𝐴 × 𝑃(𝐴)
𝑃 𝐴𝐵 =
𝑃(𝐵)

[1]
 Descriptions
Hypothesis testing vs. parameter estimation [2,3]
Debate on the process [4]
Bayes Factor (K) [3]
• Analogous to p-value or full vs reduced
model
• Way to formally compare two competing
models [5]
• Nesting not necessary
• Ratio of posterior odds for M1 to prior
odds for M1 -> so factor >> 1 supports
M1 over M2
Highest density interval analogous to
MLE with 95% CI

[6]
 Descriptions 2
Items needed for analysis [7]
• Priors
• Model
• Data
Steps for analysis
• Covered in Module III
Posterior simulation details
• Covered in Module III
 Specific tests
T-test
[8]
 Specific tests
[9]
ANOVA
 Specific tests
Regression
[10]
 Examples setup

R[11] SAS[12] SPSS[13]

Package BayesFactor PROC GENMOD Bayesian Statistics
• T-test • T-test • T-test
• ANOVA • ANOVA • ANOVA
• Simple Linear Regression • Simple Linear Regression • Simple Linear Regression
• Multiple Linear Regression • Poisson Regression

R-code available at: https://med.und.edu/daccota/_files/docs/berdc_docs/bayesian_analysis_2_r_code.txt

SAS-code available at: https://med.und.edu/daccota/_files/docs/berdc_docs/bayesian_analysis_2_sas_code.txt

 Step-by-step Example 1.1
Standard versus Bayesian Analysis in R
T-test One Sample t-test
data: diffScores
library("BayesFactor") t = -4.0621, df = 9, p-value = 0.002833
alternative hypothesis: true mean is not equal to 0
#t-test 95 percent confidence interval:
-2.4598858 -0.7001142
#Data formatting sample estimates:
data(sleep) mean of x
diffScores <- sleep$extra[1:10] - sleep$extra[11:20] -1.58

#Standard
t.test(diffScores)
Bayes factor analysis Bayes factor analysis
-------------- --------------
#Bayes [1] Alt., r=0.707 : 17.25888 ±0% [1] Null, mu=0 : 0.05794119 ±0%
bf_t <- ttestBF(x = diffScores)
bf_t #alternative model Against denominator: Against denominator:
1/bf_t #null model Null, mu = 0 Alternative, r = 0.707106781186548, mu =/= 0
--- ---
Bayes factor type: BFoneSample, JZS Bayes factor type: BFoneSample, JZS
 Step-by-step Example 1.1

T-test cont.
#Chains #Multiple
chains <-posterior(bf_t, iterations=1000) bfInterval <- ttestBF(x = diffScores, nullInterval=c(-Inf,0))
summary(chains) allbf <- c(bf_t, bfInterval)
chains2 <- recompute(chains, iterations = 10000) allbf
plot(chains2[,1:2]) plot(allbf)

Bayes factor analysis

--------------
[1] Alt., r=0.707 : 17.25888 ±0%
[2] Alt., r=0.707 -Inf<d<0 : 34.41694 ±0%
[3] Alt., r=0.707 !(-Inf<d<0) : 0.1008246 ±0.06%

Against denominator:
Null, mu = 0
---
Bayes factor type: BFoneSample, JZS
 Step-by-step Example 1.2
ANOVA
#ANOVA
#Data formatting
plot(chickwts$weight~chickwts$feed)

#Standard
summary(aov(weight~feed, data=chickwts))

#Bayes
bf_a <-anovaBF(weight~feed, data=chickwts)
bf_a Bayes factor analysis
--------------
[1] feed : 14067867 ±0%
Df Sum Sq Mean Sq F value Pr(>F)
feed 5 231129 46226 15.37 5.94e-10 *** Against denominator:
Residuals 65 195556 3009 Intercept only
--- ---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Bayes factor type: BFlinearModel, JZS
 Step-by-step Example 1.2
ANOVA cont.
#Chains
chains <-posterior(bf_a, iterations=1000)
summary(chains)
plot(chains[,2:7])

1. Empirical mean and standard deviation for each variable, 2. Quantiles for each variable:
Iterations = 1:1000
plus standard error of the mean:
Thinning interval = 1
Mean SD Naive SE Time-series SE 2.5% 25% 50% 75% 97.5%
Number of chains = 1
mu 259.35 6.588 0.20832 0.21197 mu 246.9212 254.834 259.295 263.856 272.199
Sample size per chain = 1000
feed-casein 59.04 14.627 0.46255 0.46255 feed-casein 29.0577 50.076 59.399 68.161 87.889
feed-horsebean -90.99 15.253 0.48233 0.31331 feed-horsebean -120.4751 -101.600 -90.733 -81.078 -60.935
feed-linseed -37.51 14.129 0.44680 0.47422 feed-linseed -65.4949 -47.219 -37.210 -27.900 -10.219
feed-meatmeal 17.22 15.118 0.47808 0.47808 feed-meatmeal -12.8971 7.509 17.343 27.791 45.609
feed-soybean -12.92 13.261 0.41937 0.42755 feed-soybean -39.2842 -21.363 -13.029 -4.321 13.042
feed-sunflower 65.15 14.635 0.46281 0.48797 feed-sunflower 35.7072 55.543 65.632 75.919 92.326
sig2 3138.97 571.281 18.06548 19.90381 sig2 2231.9416 2727.824 3084.627 3440.836 4512.308
g_feed 1.60 1.446 0.04572 0.04572 g_feed 0.3711 0.800 1.219 1.876 5.006
 Step-by-step Example 1.3
Simple Linear Regression
#Simple Linear Regression Call:
#Data formatting lm(formula = rating ~ complaints, data = attitude)
data(attitude)
Residuals:
par(mfrow=c(1,1))
Min 1Q Median 3Q Max
plot(x=attitude$complaints, y=attitude$rating) -12.8799 -5.9905 0.1783 6.2978 9.6294

#Standard Coefficients:
lm1 = lm(rating ~ complaints, data = attitude) Estimate Std. Error t value Pr(>|t|)
summary(lm1) (Intercept) 14.37632 6.61999 2.172 0.0385 *
complaints 0.75461 0.09753 7.737 1.99e-08 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 6.993 on 28 degrees of freedom

Multiple R-squared: 0.6813, Adjusted R-squared: 0.6699
F-statistic: 59.86 on 1 and 28 DF, p-value: 1.988e-08
 Step-by-step Example 1.3
Simple Linear Regression cont. Iterations = 1:1000
Thinning interval = 1
#Bayes Number of chains = 1
bf_r <-regressionBF(rating ~ complaints, Sample size per chain = 1000
data=attitude)
bf_r 1. Empirical mean and standard deviation for each variable,
plus standard error of the mean:
#Chains
Mean SD Naive SE Time-series SE
chains<-posterior(bf_r, iterations=1000)
mu 64.6068 1.3873 0.043871 0.04387
summary(chains) complaints 0.7286 0.1081 0.003417 0.00390
sig2 54.2648 15.4479 0.488505 0.54308
Bayes factor analysis g 6.8366 54.0517 1.709265 1.70926
--------------
[1] complaints : 417938.6 ±0.01% 2. Quantiles for each variable:

Against denominator: 2.5% 25% 50% 75% 97.5%

Intercept only mu 61.8160 63.6485 64.6576 65.5125 67.154
--- complaints 0.5051 0.6581 0.7256 0.8017 0.946
Bayes factor type: BFlinearModel, JZS sig2 31.6604 43.3656 51.5495 62.0780 91.130
g 0.2130 0.6508 1.3513 3.3668 30.291
 Step-by-step Example 1.4
Multiple Linear Regression Call:
lm(formula = rating ~ ., data = attitude)
#Multiple Linear Regression
#Data formatting Residuals:
data(attitude) Min 1Q Median 3Q Max
-10.9418 -4.3555 0.3158 5.5425 11.5990
#Standard
lm2 = lm(rating ~ ., data = attitude) Coefficients:
summary(lm2) Estimate Std. Error t value Pr(>|t|)
(Intercept) 10.78708 11.58926 0.931 0.361634
complaints 0.61319 0.16098 3.809 0.000903 ***
privileges -0.07305 0.13572 -0.538 0.595594
learning 0.32033 0.16852 1.901 0.069925 .
raises 0.08173 0.22148 0.369 0.715480
critical 0.03838 0.14700 0.261 0.796334
advance -0.21706 0.17821 -1.218 0.235577
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 7.068 on 23 degrees of freedom

Multiple R-squared: 0.7326, Adjusted R-squared: 0.6628
F-statistic: 10.5 on 6 and 23 DF, p-value: 1.24e-05
 Step-by-step Example 1.4
Multiple Linear Regression cont.
#Bayes #Compare 5 models to best
bf_r2 <-regressionBF(rating ~ ., data=attitude) bf_r2.2 <- head(bf_r2) / max(bf_r2)
length(bf_r2) 63 bf_r2.2
head(bf_r2, n=6) plot(bf_r2.2)
Bayes factor analysis Bayes factor analysis
-------------- --------------
[1] complaints : 417938.6 ±0.01% [1] complaints :1 ±0%
[2] complaints + learning : 207271.9 ±0% [2] complaints + learning : 0.4959386 ±0.01%
[3] complaints + learning + advance : 88041.54 ±0% [3] complaints + learning + advance : 0.2106566 ±0.01%
[4] complaints + raises : 77498.99 ±0% [4] complaints + raises : 0.1854315 ±0.01%
[5] complaints + privileges : 75015.23 ±0% [5] complaints + privileges : 0.1794886 ±0.01%
[6] complaints + advance : 72759.76 ±0% [6] complaints + advance : 0.174092 ±0.01%

Against denominator: Against denominator:

Intercept only rating ~ complaints
--- ---
Bayes factor type: BFlinearModel, JZS Bayes factor type: BFlinearModel, JZS
 Assessment 1

https://und.qualtrics.com/jfe/form/SV_eW1CNrs7cPhjqFE
 Step-by-step Example 2.1
Standard versus Bayesian Analysis in SAS
T-test
DATA Iris; set sashelp.Iris;
PROC FREQ data=Iris;
tables Species;
PROC MEANS data=Iris;
where Species in ("Setosa" "Versicolor");
var SepalLength;
by Species;

PROC SGPLOT data=Iris;

where Species in ("Setosa" "Versicolor"); Iris Species=Setosa

histogram SepalLength /group=Species transparency=0.30; Analysis Variable : SepalLength Sepal Length (mm)
N Mean Std Dev Minimum Maximum
Iris Species
50 50.0600000 3.5248969 43.0000000 58.0000000
Cumulative Cumulative
Species Frequency Percent Frequency Percent Iris Species=Versicolor
Setosa 50 33.33 50 33.33 Analysis Variable : SepalLength Sepal Length (mm)
Versicolor 50 33.33 100 66.67 N Mean Std Dev Minimum Maximum
Virginica 50 33.33 150 100.00 50 59.3600000 5.1617115 49.0000000 70.0000000
 Step-by-step Example 2.1
Analysis Of Maximum Likelihood Parameter Estimates
T-test cont. Standard Wald 95% Wald Chi- Pr > ChiS
PROC GENMOD data=Iris; Parameter DF Estimate Error Confidence Limits Square q
Intercept 1 59.3600 0.6188 58.1472 60.5728 9203.20 <.0001
where Species in ("Setosa" "Versicolor");
Species Setosa 1 -9.3000 0.8751 -11.0151 -7.5849 112.95 <.0001
class Species;
Species Versicolor 0 0.0000 0.0000 0.0000 0.0000 . .
model SepalLength=Species;
Scale 1 4.3753 0.3094 3.8091 5.0257

PROC GENMOD data=Iris; Posterior Summaries

where Species in ("Setosa" "Versicolor"); Standard Percentiles
class Species; Parameter N Mean Deviation 25% 50% 75%
model SepalLength=Species; Intercept 10000 59.3613 0.6312 58.9320 59.3711 59.7808
bayes seed=1 coeffprior=normal; SpeciesSetosa 10000 -9.3018 0.8966 -9.9028 -9.2927 -8.6954
Dispersion 10000 19.9560 2.9351 17.8867 19.6865 21.7329

Posterior Intervals
Parameter Alpha Equal-Tail Interval HPD Interval
Intercept 0.050 58.1180 60.6000 58.1440 60.6221
SpeciesSetosa 0.050 -11.0525 -7.5646 -11.1072 -7.6300
Dispersion 0.050 15.0091 26.5130 14.8495 26.0907
 Step-by-step Example 2.1

T-test cont. 2
PROC GENMOD data=Iris;
where Species in ("Setosa" "Versicolor");
class Species;
model SepalLength=Species;
bayes seed=1 coeffprior=normal;
 Step-by-step Example 2.2
ANOVA Analysis Of Maximum Likelihood Parameter Estimates
PROC SGPLOT data=Iris; Standard Wald 95% Wald Chi- Pr > ChiS
Parameter DF Estimate Error Confidence Limits Square q
vbox SepalLength /group=Species;
Intercept 1 65.8800 0.7207 64.4674 67.2926 8355.87 <.0001
Species Setosa 1 -15.8200 1.0192 -17.8177 -13.8223 240.92 <.0001
PROC GENMOD data=Iris;
Species Versicolor 1 -6.5200 1.0192 -8.5177 -4.5223 40.92 <.0001
class Species;
Species Virginica 0 0.0000 0.0000 0.0000 0.0000 . .
model SepalLength=Species;
Scale 1 5.0962 0.2942 4.5509 5.7067

PROC GENMOD data=Iris; Posterior Summaries

class Species; Standard Percentiles
model SepalLength=Species; Parameter N Mean Deviation 25% 50% 75%
bayes seed=1 coeffprior=uniform Intercept 10000 65.8905 0.7270 65.4044 65.8926 66.3677
SpeciesSetosa 10000 -15.8376 1.0319 -16.5273 -15.8303 -15.1419
SpeciesVersicolor 10000 -6.5454 1.0245 -7.2170 -6.5473 -5.8728
Dispersion 10000 26.8852 3.1557 24.6744 26.6780 28.8562

Posterior Intervals
Parameter Alpha Equal-Tail Interval HPD Interval
Intercept 0.050 64.4771 67.3561 64.4578 67.3284
SpeciesSetosa 0.050 -17.8639 -13.8305 -17.9082 -13.8751
SpeciesVersicolor 0.050 -8.5717 -4.5105 -8.6151 -4.5645
Dispersion 0.050 21.3747 33.7054 20.9688 33.1140
 Step-by-step Example 2.2
ANOVA cont.
PROC GENMOD data=Iris;
class Species;
model SepalLength=Species;
bayes seed=1 coeffprior=uniform;
 Step-by-step Example 2.3
Linear Regression
PROC GLIMMIX data=Liver;
model LC_nodes = BMI Age Time BCM AHBA Jaund / dist=Normal;

PROC GENMOD data=Liver;

Criteria For Assessing Goodness Of Fit
model LC_nodes = BMI Age Time BCM AHBA Jaund / dist=Normal;
Criterion DF Value Value/DF
Pearson Chi-Square 4006.57 Deviance 129 4006.5654 31.0586
Scaled Deviance 129 136.0000 1.0543
Pearson Chi-Square / DF 31.06
Pearson Chi-Square 129 4006.5654 31.0586
Type III Tests of Fixed Effects
Effect Num DF Den DF F Value Pr > F Analysis Of Maximum Likelihood Parameter Estimates
BMI 1 129 0.04 0.8478 Standard Wald 95% Wald Chi- Pr > ChiS
Age 1 129 5.81 0.0173 Parameter DF Estimate Error Confidence Limits Square q
Time 1 129 0.37 0.5423 Intercept 1 10.0684 2.8684 4.4464 15.6904 12.32 0.0004
BCM 1 129 2.01 0.1585 BMI 1 -0.0180 0.0912 -0.1968 0.1608 0.04 0.8435
AHBA 1 129 2.59 0.1103 Age 1 -0.0779 0.0315 -0.1396 -0.0162 6.13 0.0133
Jaund 1 129 0.97 0.3268 Time 1 -0.0167 0.0267 -0.0690 0.0355 0.39 0.5304
BCM 1 -1.4874 1.0213 -3.4892 0.5144 2.12 0.1453
AHBA 1 1.8310 1.1089 -0.3424 4.0043 2.73 0.0987
Jaund 1 0.9864 0.9761 -0.9266 2.8995 1.02 0.3122
Scale 1 5.4277 0.3291 4.8195 6.1126
 Step-by-step Example 2.3
Linear Regression cont.
PROC GENMOD data=Liver;
model LC_nodes = BMI Age Time BCM AHBA Jaund / dist=Normal;
bayes seed=1 coeffprior=normal;
Posterior Intervals
Posterior Summaries
Parameter Alpha Equal-Tail Interval HPD Interval
Percentiles Intercept 0.050 4.1960 15.9416 4.4029 16.1206
Standard
Parameter N Mean Deviation 25% 50% 75% BMI 0.050 -0.2038 0.1698 -0.2066 0.1663
Intercept 10000 10.0877 3.0033 8.0747 10.1044 12.1463 Age 0.050 -0.1414 -0.0134 -0.1384 -0.0110
Time 0.050 -0.0707 0.0381 -0.0713 0.0371
BMI 10000 -0.0188 0.0957 -0.0839 -0.0183 0.0454
BCM 0.050 -3.5754 0.5938 -3.5908 0.5647
Age 10000 -0.0776 0.0323 -0.0994 -0.0776 -0.0561 AHBA 0.050 -0.4079 4.0623 -0.3635 4.0957

Time 10000 -0.0167 0.0277 -0.0352 -0.0168 0.00192 Jaund 0.050 -0.9916 2.9613 -1.0089 2.9212
Dispersion 0.050 24.6876 40.2912 24.1990 39.6035
BCM 10000 -1.5052 1.0603 -2.2129 -1.5099 -0.7910

AHBA 10000 1.8320 1.1369 1.0794 1.8338 2.5821

Jaund 10000 0.9794 1.0128 0.2964 0.9762 1.6622

Dispersion 10000 31.5808 4.0150 28.7134 31.2603 34.0755

 Step-by-step Example 2.3
Linear Regression cont. 2
PROC GENMOD data=Liver;
model LC_nodes = BMI Age Time BCM AHBA Jaund / dist=Normal;
bayes seed=1 coeffprior=normal;
 Step-by-step Example 2.4
Poisson Regression
PROC GLIMMIX data=Liver;
model LC_nodes = BMI Age Time BCM AHBA Jaund / dist=Poisson;

PROC GENMOD data=Liver; Criteria For Assessing Goodness Of Fit

model LC_nodes = BMI Age Time BCM AHBA Jaund / dist=Poisson link=log; Criterion DF Value Value/DF
Deviance 129 382.1373 2.9623
Pearson Chi-Square 501.12 Scaled Deviance 129 382.1373 2.9623
Pearson Chi-Square / DF 3.88 Pearson Chi-Square 129 501.1176 3.8846

Type III Tests of Fixed Effects Analysis Of Maximum Likelihood Parameter Estimates
Effect Num DF Den DF F Value Pr > F Standard Wald 95% Confidence Wald Chi- Pr > ChiS
BMI 1 129 0.29 0.5889 Parameter DF Estimate Error Limits Square q
Age 1 129 32.34 <.0001 Intercept 1 2.4508 0.2284 2.0032 2.8984 115.16 <.0001
Time 1 129 1.79 0.1833 BMI 1 -0.0044 0.0080 -0.0201 0.0114 0.29 0.5880
BCM 1 129 11.67 0.0008 Age 1 -0.0135 0.0024 -0.0181 -0.0088 32.34 <.0001
AHBA 1 129 14.94 0.0002 Time 1 -0.0029 0.0022 -0.0072 0.0014 1.79 0.1810
Jaund 1 129 6.31 0.0132 BCM 1 -0.2715 0.0795 -0.4272 -0.1157 11.67 0.0006
AHBA 1 0.3215 0.0832 0.1585 0.4845 14.94 0.0001
Jaund 1 0.2077 0.0827 0.0456 0.3698 6.31 0.0120
Scale 0 1.0000 0.0000 1.0000 1.0000
 Step-by-step Example 2.4
Poisson Regression cont.
PROC GENMOD data=Liver;
model LC_nodes = BMI Age Time BCM AHBA Jaund / dist=Poisson link=log;
bayes seed=1 plots=none coeffprior=normal;

Posterior Summaries Posterior Intervals

Parameter Alpha Equal-Tail Interval HPD Interval
Percentiles Intercept 0.050 1.9903 2.9059 2.0289 2.9321
Standard
Parameter N Mean Deviation 25% 50% 75% BMI 0.050 -0.0209 0.0108 -0.0211 0.0106
Intercept 10000 2.4483 0.2320 2.2903 2.4493 2.6093 Age 0.050 -0.0181 -0.00870 -0.0184 -0.00908
Time 0.050 -0.00761 0.00105 -0.00745 0.00113
BMI 10000 -0.00475 0.00809 -0.0101 -0.00466 0.000851
BCM 0.050 -0.4257 -0.1063 -0.4314 -0.1152
Age 10000 -0.0134 0.00237 -0.0150 -0.0134 -0.0118 AHBA 0.050 0.1563 0.4804 0.1574 0.4811
Time 10000 -0.00303 0.00220 -0.00445 -0.00298 -0.00150 Jaund 0.050 0.0450 0.3777 0.0468 0.3788

BCM 10000 -0.2703 0.0799 -0.3241 -0.2725 -0.2190

AHBA 10000 0.3202 0.0828 0.2642 0.3209 0.3775

Jaund 10000 0.2106 0.0838 0.1533 0.2111 0.2663

 Step-by-step Example 2.4
Poisson Regression cont. 2
 Step-by-step Example 3.1
Standard versus Bayesian Analysis in SPSS
T-test
 Step-by-step Example 3.1
T-test cont.
 Step-by-step Example 3.1
T-test cont. 2

[14]
 Step-by-step Example 3.2
ANOVA
 Step-by-step Example 3.2
ANOVA cont.

*
 Step-by-step Example 3.3
Simple Linear Regression
 Step-by-step Example 3.3
Simple Linear Regression cont.
 Assessment 2

https://und.qualtrics.com/jfe/form/SV_0ufeUVznLiqeZJc
 Caveats and Concerns

• Not always a strong difference between Frequentist and Bayesian

• Most of what we’ve covered is relatively trivial
• Move from familiarity to unfamiliarity
• Choice of priors can be important (Module III)
• Need to check fit of posterior modeling (Module III)
 Real World Examples
[15]
Brard C, Le Teuff G, Le Deley MC, Hampson LV. Bayesian survival
analysis in clinical trials: What methods are used in practice? Clin
Trials. 2017 Feb;14(1):78-87. doi: 10.1177/1740774516673362.
Epub 2016 Oct 11. PMID: 27729499.
A systematic review of clinical trials using Bayesian survival analyses was
performed through PubMed and Web
of Science databases.

In total, 28 articles met the inclusion criteria, 25 were original reports of

clinical trials and 3 were re-analyses of a clinical trial.

Few trials implemented a Bayesian survival analysis and few incorporated

external data into priors.
 Real World Examples
[16]
Bittl JA, He Y. Bayesian Analysis: A Practical Approach to Interpret
Clinical Trials and Create Clinical Practice Guidelines. Circ
Cardiovasc Qual Outcomes. 2017;10(8):e003563.
doi:10.1161/CIRCOUTCOMES.117.003563
In this review, we present gradually more complex examples,
along with programming code and data sets, to show how
Bayesian analysis takes evidence from randomized clinical trials to
update what is already known about specific treatments in
cardiovascular medicine.
To perform traditional meta-analyses, we use the open-source
statistical program R and library package meta. To generate
conjugate-normal models, we combine normal probability
distributions from older trial data (prior) and new trial results
(likelihood) to generate the posterior. To perform more complex
computations, we use a version of BUGS called OpenBUGS that
allows Markov chain Monte Carlo modeling to specify the
posterior distribution.
 Real World Examples
[17]
Luo C, Wang L, Wu G, et al. Comparison of the efficacy of hematopoietic
stem cell mobilization regimens: a systematic review and network meta-
analysis of preclinical studies. Stem Cell Res Ther. 2021;12(1):310.
Published 2021 May 29. doi:10.1186/s13287-021-02379-6

Bayesian network meta-analyses were performed following the guidelines of the

National Institute for Health and Care Excellence Decision Support Unit (NICE
DSU) with WinBUGS version 1.4.3.
 Real World Examples
[18]
Price MN, Dehal PS, Arkin AP (2010) FastTree 2 – Approximately Maximum-Likelihood Trees for Large Alignments. PLoS ONE 5(3):
e9490. doi:10.1371/ journal.pone.0009490
FastTree recomputes all posterior distributions and recalculates the log likelihood of the tree at each site. FastTree then uses a Bayesian approach to select
which rate to use at each site: FastTree maximizes P(rate|Site) ∝ Lk(Site|rate) ς (rate) where ς (rate) is a gamma-distributed prior.
 Summary and Conclusion
• Bayesian analysis takes in prior beliefs (hypothesis) and evidence
(data) to generate posterior beliefs (probability)
• The two major methods of using Bayesian analysis is 1) the Bayes
factor for hypothesis testing (p-value analog) and 2) highest density
interval for estimation with uncertainty (ML analog)
• Bayesian analysis can be run in R, SAS, and SPSS
• Tune in next time for deeper details into Bayesian analysis in
Bayesian Analysis Module III: Deep Dive
 References
[1] https://www.digitalvidya.com/wp-content/w3-webp/uploads/2018/11/Bayesian_inference_archaeology_example.jpgw3.webp
[2] https://doingbayesiandataanalysis.blogspot.com/2016/12/bayesian-assessment-of-null-values.html
[3] https://link.springer.com/content/pdf/10.3758/s13423-016-1221-4.pdf
[4] https://stats.stackexchange.com/questions/246481/bayesian-parameter-estimation-or-bayesian-hypothesis-testing
[5] https://people.stat.sc.edu/Hitchcock/stat535slidesday20.pdf
[6] https://en.wikipedia.org/wiki/Bayes_factor
[7] https://www.slideshare.net/ASQwebinars/general-bayesian-methods-for-typical-reliability-data-analysis
[8] https://m-clark.github.io/models-by-example/bayesian-t-test.html
[9] https://www.cairn.info/revue-l-annee-psychologique-2020-1-page-73.htm
[10] https://gregorygundersen.com/blog/2020/02/04/bayesian-linear-regression/
[11] https://cran.r-project.org/web/packages/BayesFactor/vignettes/manual.html
[12] https://support.sas.com/resources/papers/proceedings14/SAS400-2014.pdf
[13] https://www.ibm.com/docs/en/spss-statistics/25.0.0?topic=statistics-bayesian
[14] https://www.ibm.com/docs/en/spss-statistics/25.0.0?topic=statistics-bayesian-independent-sample-inference
[15] https://pubmed.ncbi.nlm.nih.gov/27729499/
[16] https://pubmed.ncbi.nlm.nih.gov/28798016/
[17] https://stemcellres.biomedcentral.com/track/pdf/10.1186/s13287-021-02379-6.pdf
[18] https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0009490
 Acknowledgements
• The DaCCoTA is supported by the National
Institute of General Medical Sciences of the
National Institutes of Health under Award
Number U54GM128729.
• For the labs that use the Biostatistics,
Epidemiology, and Research Design Core in
any way, including this Module, please
acknowledge us for publications. "Research
reported in this publication was supported
by DaCCoTA (the National Institute of
General Medical Sciences of the National
Institutes of Health under Award Number
U54GM128729)".