Scribd
Upload
46 views

Statistical Methods For Cross-Sectional Data Analysis

This document discusses various statistical methods for analyzing cross-sectional data, including: 1) Regression methods such as ordinary least squares (OLS), generalized least squares (GLS), and weighted least squares (WLS) for different data distributions and heterogeneity. 2) Robust regression approaches that are robust to outliers like regression with robust standard errors. 3) Methods for censored, truncated, and measurement error data like regression with censored or truncated data. 4) Other approaches involving multiple variables like seemingly unrelated regression, multivariate regression, and structural equation modeling.

Uploaded by

Queen A
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
46 views

Statistical Methods For Cross-Sectional Data Analysis

This document discusses various statistical methods for analyzing cross-sectional data, including: 1) Regression methods such as ordinary least squares (OLS), generalized least squares (GLS), and weighted least squares (WLS) for different data distributions and heterogeneity. 2) Robust regression approaches that are robust to outliers like regression with robust standard errors. 3) Methods for censored, truncated, and measurement error data like regression with censored or truncated data. 4) Other approaches involving multiple variables like seemingly unrelated regression, multivariate regression, and structural equation modeling.

Uploaded by

Queen A
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 1

Statistical Methods for Cross-Sectional Data Analysis

Regression Cross-sectional data


If data follow normal distribution and no have heterogeneity
OLS
problem
If data do not follow normal distribution and have
GLS
heterogeneity problem. It’s good for primary data
If data do not follow normal distribution and have
WLS
heterogeneity problem. It’s good for secondary data
Three approaches to robust regression: 1) regression with
robust standard errors including the cluster option, 2) robust
Robust
regression using iteratively reweighted least squares, and 3)
quantile regression, more specifically, median regression.
Constrained Linear
where constrain coefficients to be equal to each other
Regression
A censored value is one whose value is incomplete due to
Regression with
random factors for each subject. A truncated observation, on
Censored or Truncated
the other hand, is one which is incomplete due to a selection
Data
process in the design of the study.
Regression with one of the assumptions of regression is that the predictor
Measurement Error variables are measured without error
Seemingly Unrelated It is the case that the errors (residuals) from these two models
Regression would be correlated
More than one DV and run together. It is same as Seemingly
Multivariate Regression
Unrelated Regression
Binary Logistic When DV is binary, IV can be any type. This can be run
Regression under logit, probit and tobit model
When DV is likert scale (order form), IV can be any type.
Ordinal Regression This can be run under logit, probit, tobit mode, multinomial,
and multilevel model.
Path diagram (measurement Model) measure the relationship
among contribution of different observable variable as well
Structural Equation as latent variable. If sample size is small (>300), better to use
Model PLS. If sample size is big (<300), better to use AMOS. For
both cases, can use STATA software. It is also used for
Factor Analysis.
To check the variance are same for two DV. Here the data of
ANOVA, MANOVA
factors variable should be in categorical variables
To check the mean are same for two variables. Here the data
Χ2 Test
should be in likert scale (order form)
To check the mean are same for two variables. Here the data
t-test
should be in ratio scale
Data can be check by find out the number of group or
Grouping
category available, can be check by RASCH software
When the pattern of the observation can be understand with
Simulation (Monte different probability value and huge number of trails (5-10
Carlo) thousands). In that case, mean and standard deviation of
actual data is needed to run trials.

Prepared by - Mahadi Hasan; Md. Mahmudul Alam


* You are free to develop the file & circulate freely. Please notify us if there is any improvement.

You might also like