Scribd
Upload
119 views

Bayesian Data Analysis: Introduction

This document discusses the conceptual framework of Bayesian data analysis. It explains that Bayesian inference involves updating the probability or credibility of hypotheses as new evidence becomes available, based on Bayes' theorem. The key steps are: (1) defining possible parameter values in descriptive models to represent the data; (2) specifying a prior distribution over the parameter values; (3) using Bayesian inference to redistribute credibility across parameter values based on the observed data; and (4) checking if the resulting posterior predictions fit the data well. Parameterized models are necessary for Bayesian analysis, even if they have an infinite number of parameters as in nonparametric models.

Uploaded by

Sohail Tarar
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
119 views

Bayesian Data Analysis: Introduction

This document discusses the conceptual framework of Bayesian data analysis. It explains that Bayesian inference involves updating the probability or credibility of hypotheses as new evidence becomes available, based on Bayes' theorem. The key steps are: (1) defining possible parameter values in descriptive models to represent the data; (2) specifying a prior distribution over the parameter values; (3) using Bayesian inference to redistribute credibility across parameter values based on the observed data; and (4) checking if the resulting posterior predictions fit the data well. Parameterized models are necessary for Bayesian analysis, even if they have an infinite number of parameters as in nonparametric models.

Uploaded by

Sohail Tarar
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 13

Sohail Manzoor

PhD Student
Division of Mining and Geotechnical Engineering, Luleå University of Technology, Sweden
1. Bayesian Inference Is Reallocation of Credibility Across Possibilities
1.1 Data are noisy and inferences are probabilistic

2. Possibilities Are Parameter Values in Descriptive Models

3. The Steps of Bayesian Data Analysis


3.1 Data analysis without parametric models?
1. Bayesian Inference Is Reallocation of Credibility Across Possibilities
1.1 Data are noisy and inferences are probabilistic

2. Possibilities Are Parameter Values in Descriptive Models

3. The Steps of Bayesian Data Analysis


3.1 Data analysis without parametric models?

• Goal: To introduce the conceptual framework of Bayesian data analysis.

1. The first idea is that Bayesian inference is reallocation of credibility across possibilities.
2. The second foundational idea is that the possibilities, over which we allocate credibility,
are parameter values in meaningful mathematical models.
Bayesian inference = Statistical inference + Bayes' theorem

• Statistical inference is the process of using data analysis to deduce properties of an underlying
probability distribution (probabilities of occurrence of different possible outcomes in an
experiment).

• Bayes’ theorem (alternatively Bayes’ law or Bayes’ rule) describes the probability of a possible
outcome, based on prior knowledge of conditions that might be related to the outcome.

• Bayesian inference is used to update the probability (credibility) for a hypothesis as more
evidence or information becomes available.
• In reality, data have only probabilistic relations to their underlying causes.

• All scientific data have some degree of “noise” in their values.

• We can collect data and only incrementally adjust the credibility of some possible trends.

• The beauty of Bayesian analysis is that the mathematics reveal exactly how much to re-allocate
credibility in realistic probabilistic situations.
• A key step in Bayesian analysis is defining the set of possibilities over which credibility is allocated.

• Posterior predictive check.

• In general, data analysis begins with a family of candidate descriptions for the data.

• In realistic applications, the candidate parameter values can form an infinite continuum, not only a
few discrete options.

• There are two main requirements for a mathematical description of data.


1. First, the mathematical form should be comprehensible with meaningful parameters.
2. Second, it should be descriptively adequate, which means, loosely, that the mathematical
form should “look like” the data.

• It is also important to understand that mathematical descriptions of data are not necessarily
causal explanations of data.
• In general, Bayesian analysis of data follows these steps:

1. Identify the data relevant to the research questions. What are the measurement scales of the data?
Which data variables are to be predicted, and which data variables are supposed to act as predictors?

2. Define a descriptive model for the relevant data. The mathematical form and its parameters
should be meaningful and appropriate to the theoretical purposes of the analysis.

3. Specify a prior distribution on the parameters. The prior must pass muster with the audience of the
analysis, such as skeptical scientists.

4. Use Bayesian inference to re-allocate credibility across parameter values. Interpret the posterior
distribution with respect to theoretically meaningful issues (assuming that the model is a
reasonable description of the data; see next step).

5. Check that the posterior predictions mimic the data with reasonable accuracy (i.e., conduct a
“posterior predictive check”). If not, then consider a different descriptive model.
• Suppose, we are interested in the relationship between weight and height of people.
• We suspect from everyday experience that taller people tend to weigh more than shorter people.
• But we would like to know
1. how much people’s weights tend to increase when height increases, and
2. how certain we can be about the magnitude of the increase.
• In particular, we might be interested in predicting a person’s weight based on their height.

• The first step is identifying the relevant data.


• The second step is to define a descriptive model of the data that is meaningful.
ˆy = β1x + β0
y ∼ normal(ˆy, σ)
• The third step in the analysis is specifying a prior distribution on the parameters.
• The fourth step is interpreting the posterior distribution. (different aspects)
• The fifth step is to check that the model, with its most credible parameter values, actually mimics
the data reasonably well. This is called a “posterior predictive check.”
• One situation in which it might appear that parameterized models are not used is with so-called
nonparametric models.

• But these models are confusingly named because they actually do have parameters; in fact they
have a potentially infinite number of parameters.

• We will not be considering Bayesian nonparametric models in this book.

• Bayesian computations for the situations when the parameters refer to discrete states instead of
continuous distributions are discussed in Chapter 5.

• Finally, there might be some situations in which the analyst hate to commit to any parameterized
model of the data. If this is the case, then Bayesian methods cannot apply.
• Model A: P(x) = ¼
• Model B: P(x) = x/10
P(1) = 1/10
P(2) = 2/10 = 1/5
P(3) = 3/10
P(4) = 4/10 = 2/5
• Model C: P(x) = 12/25x
P(1) = 12/25
P(2) = 12/50 = 6/25
P(3) = 12/75 = 4/25
P(4) = 12/100 = 3/25
• Model A
P(x) = ¼

• Model C
P(x) = 12/25x
Address:
Division of Mining & Geotechnical Engineering
Luleå University of Technology, Luleå, Sweden

Email:
[email protected]

THANK Contact:

YOU +46 72 22 895 44


+46 92 04 914 18

You might also like