Presented by

Confirmatory Factor
Analysis
Confirmatory Factor Analysis
(CFA)
Is a statistical technique used primarily in social sciences to test whether
a set of observed variables aligns with a hypothesized factor structure.
This method is distinct from Exploratory Factor Analysis (EFA), which is
used to discover the underlying relationships without a predefined
model. CFA is employed to validate theories regarding the relationships
between observed variables and their latent constructs.
 Objectives and Process of CFA
The main goal of CFA is to confirm if the data fits a proposed measurement
model based on prior theoretical frameworks or empirical research. The process
typically involves several critical steps:
01 02 03 04 05
Defining Constructs- Developing the Specifying the Assessing Model Fit: The Considerations:
Researchers begin Measurement Model: fit of the measurement Successful CFA
by clearly defining Model: Researchers model is evaluated using requires adherence to
the theoretical Establishing must specify various indices, such as certain assumptions,
constructs they unidimensionality the number of Chi-square, Root Mean including multivariate
wish to measure. is essential, where factors and the Square Error of normality, an
This often includes each factor is loading Approximation (RMSEA), adequate sample size
pretesting to ensure represented by patterns of the Goodness of Fit Index (usually n > 200), and
that the multiple observed observed (GFI), and Comparative Fit proper model
measurement items variables. A variables on Index (CFI). Adequate specification based on
accurately reflect common practice these factors, factor loadings (typically theoretical or
the intended is to have at least based on above 0.7) are also empirical justification.
constructs. three items per theoretical assessed to determine Random sampling is
construct. expectations. the validity of the model. also crucial to ensure
that findings can be
generalized.
How does CFA differ from Exploratory
Factor Analysis (EFA)
Confirmatory Factor Analysis (CFA) and Exploratory Factor Analysis (EFA) are both
techniques used in factor analysis, but they serve different purposes and are applied in
distinct contexts.
• Purpose: • Assumptions:
EFA: Used to explore the underlying EFA: Assumes that the researcher
structure of a set of variables without has little or no prior knowledge about
any preconceived notions about how the number of factors or the
many factors exist or how they relate relationships between variables. It
to one another. It is primarily a data- allows variables to load on multiple
driven approach aimed at identifying factors.
potential factor structures. CFA: Requires a clear theoretical
CFA: Used to test a specific framework or prior empirical evidence
hypothesized factor structure based on to specify the number of factors and
prior research or theory. It assesses the expected loadings of variables on
whether the observed data fits the those factors. Each variable is assigned
predefined model, confirming whether to a specific factor.
the data aligns with the expected
relationships among variables.
 Other reasons..
• Model Specification: • Output and Interpretation:

EFA: The number of factors is EFA: Produces a factor solution that

determined during the analysis, and can vary with different datasets,
the model is flexible, allowing for the providing insights into possible
exploration of different structures. structures.
CFA: The number of factors and their CFA: Produces fit indices (like RMSEA,
relationships are specified a priori, and CFI) that indicate how well the
the analysis tests how well the data fits hypothesized model fits the data,
this model. allowing for hypothesis testing.
 When to Use Each Method
Use EFA when: Use CFA when:
• You have no clear theoretical • You have a well-defined theoretical
framework or prior research to model or prior research supporting
guide your expectations about the a specific factor structure.
factor structure. • You want to confirm hypotheses
• You aim to identify the underlying about the relationships between
factors in a new measure or observed variables and their latent
dataset. constructs.

In practice, researchers often use EFA in the initial stages of

instrument development to explore potential factor structures, followed by
CFA to validate the identified structure with a different sample
 When to Choose EFA
Lack of Prior Theory: Use EFA when New Measurement Instruments: If you
there is no established theoretical are developing a new questionnaire or
framework or prior research guiding measurement tool, EFA is
your expectations about the factor recommended to explore the
structure. EFA is ideal for exploring dimensions of the data and determine
data to identify potential underlying how many factors are present. This
factors without preconceived notions helps in refining the instrument before
testing it further with CFA
Identifying Factor Structure: EFA is Data Exploration: If your goal is to
appropriate when you want to discover reduce dimensionality or to understand
how many factors can be extracted the structure of complex data sets, EFA
from a set of variables and how these provides a method to uncover
factors relate to each other. It allows relationships among variables without
for flexibility in determining the the constraints of a predefined model
number of factors and their
relationships

In summary, EFA is best utilized in the early stages of research when exploring
new constructs or developing measurement tools, while CFA is suited for testing
specific hypotheses about the relationships between observed variables and their
underlying factors. Using EFA first can provide valuable insights that inform
subsequent CFA, ensuring a robust analysis of the data.
 When to Choose CFA
Hypothesis Testing: CFA is appropriate Theoretical Framework: If your
when you have a clear hypothesis research is grounded in existing theory
regarding the relationships between or prior empirical research that
observed variables and their suggests a specific factor structure,
underlying latent constructs. This CFA allows you to confirm whether that
contrasts with Exploratory Factor structure holds true in your data
Analysis (EFA), which is used to
discover the factor structure without
predefined hypotheses

Measurement Model Validation: CFA is Model Specification: Before conducting

ideal for validating measurement CFA, researchers must specify the
models, especially when developing number of factors and the expected
surveys or scales. It helps ensure that loadings of observed variables on
the items intended to measure specific these factors based on theoretical
constructs do so effectively, confirming expectations. This requires a well-
the unidimensionality of each factor defined measurement model
 Other reasons..

Assessing Model Fit: CFA provides Sample Size Considerations: Adequate

various fit indices (e.g., Chi-square, sample size is essential for CFA,
RMSEA, CFI) to assess how well the typically requiring at least 200
proposed model fits the data. This is respondents to ensure reliable results.
crucial for determining the validity of This is particularly important when
the measurement model estimating multiple parameters in the
model

Multivariate Normality: The data should

ideally follow a multivariate normal
distribution, which is a prerequisite for
conducting CFA effectively
 Determinants in using CFA
Sample Size Missing Data Normality
• CFA requires an • Missing data can • Univariate and
adequate sample size, significantly impact multivariate normality are
typically at least 200 CFA results. Assess the important assumptions
respondents for extent and pattern of for CFA.
reliable results. missing data in your • Check for univariate
• The Satorra–Saris dataset. normality by assessing
method, MacCallum • Use appropriate skewness and kurtosis of
approach, and Monte techniques to handle individual variables.
Carlo approach can missing data, such as • Evaluate multivariate
help determine the imputation or normality using Mardia's
minimum sample size maximum likelihood coefficient or other
needed for your estimation, depending appropriate tests.
specific mode on the amount and • If data is non-normal,
type of missing data consider using robust
estimation methods like
maximum likelihood with
robust standard errors or
asymptotically
distribution-free
estimation
 Other determinants
Levels of Measurement Linearity and Multicollinearity Model Specification
• CFA assumes that • CFA assumes linear • Specify your
observed variables are relationships between hypothesized factor
measured on at least an observed variables and structure based on
ordinal scale. latent factors. theory and prior
• Ensure that your • Check for multicollinearity research.
variables meet this among observed variables, • Ensure that the model is
requirement based on as high correlations can identified, meaning
the measurement scale lead to estimation issues there are enough known
used in your study parameters to estimate
the unknown
parameters
 Handle missing data in my dataset for CFA

1. Understanding Missing Data Mechanisms

Before addressing missing data, it's important to understand why
data is missing. The three main types are:
• Missing Completely at Random (MCAR): The missingness is unrelated
to any observed or unobserved data. If data is MCAR, you can safely
remove cases with missing values without introducing bias.
• Missing at Random (MAR): The missingness is related to observed
data but not the missing data itself. In this case, you can use
techniques like imputation to handle the missing values.
• Missing Not at Random (MNAR): The missingness is related to the
unobserved data. This scenario is more complex and may require
specialized methods to address.
2. Assessing Missing Data
• To determine the nature of your missing data, you can:Create dummy
variables indicating whether values are missing (1 for missing, 0 for
observed) and run statistical tests (e.g., t-tests for continuous variables) to
see if the missingness correlates with other variables. Significant results
suggest the data may be MAR rather than MCAR

3. Imputation Methods
If your data is MAR, consider the following imputation techniques:
• Full Information Maximum Likelihood (FIML): This method uses all available
data to estimate parameters and is valid under the MAR assumption. Many
software packages, including JASP and lavaan, implement FIML
automatically for CFA.
• Mean/Median Imputation: Substitute missing values with the mean or
median of the observed values. While simple, this method can reduce
variability and may not be suitable if the percentage of missing data is
high.
• Multiple Imputation: This technique creates several different plausible
datasets by imputing missing values multiple times, then combines the
results. This approach accounts for the uncertainty of the missing data.
• Last Observation Carried Forward (LOCF): In longitudinal data, replace
missing values with the last observed value. This method is straightforward
but can introduce bias if the data has a trend
4. Data Removal
If the percentage of missing data is small, you might consider removing cases
with missing values. However, be cautious, as this can lead to loss of valuable
information and reduce the sample size significantly.

5. Sensitivity Analysis
Conduct sensitivity analyses by comparing results from the complete cases
(those without any missing data) against those from the full dataset using
imputation methods. This can help assess the robustness of your findings
Applications and Software
CFA is widely used in the development and validation of measurement
instruments, such as surveys and psychological scales. It helps
researchers ascertain whether their instruments effectively measure the
constructs they are intended to assess. Common statistical software for
conducting CFA includes AMOS, LISREL, and SAS, which offer various
tools for model specification and analysis.In summary, CFA is an
essential tool for researchers looking to validate their measurement
models and ensure the reliability and validity of their findings, making it
a critical component in the toolkit of quantitative research
methodologies