Theme: Statistics

This course will provide a detailed critique of the methods and philosophy of the Null Hypothesis Significance Testing (NHST) approach to statistics which is currently dominant in social and biomedical science. We will briefly contrast NHST with alternatives, especially with Bayesian methods. We will use some computer code (Matlab and R) to demonstrate some issues. However, we will focus on the big picture rather on the implementation of specific procedures.

The purpose of this course is to familiarise students with the basic concepts of Bayesian theory. It is designed to provide an introduction to the principles, methods, and applications of Bayesian statistics. Bayesian statistics offers a powerful framework for data analysis and inference, allowing for the incorporation of prior knowledge and uncertainty in a coherent and systematic manner.

Throughout this course, we will cover key concepts such as Bayes' theorem, prior and posterior distributions, likelihood functions, and the fundamental differences between Bayesian and frequentist approaches. You will learn to formulate and estimate statistical models, update beliefs using new data, and make informed decisions based on the posterior probabilities generated through Bayesian inference. By the end of this course, you will possess the necessary skills to perform Bayesian data analysis, interpret results, and apply Bayesian methods in various contexts.

This course aims to provide students with a range of specific technical skills that will enable them to undertake impact evaluation of policy. Too often policy is implemented but not fully evaluated. Without evaluation we cannot then tell what the short or longer term impact of a particular policy has been. On this course, students will learn the skills needed to evaluate particular policies and will have the opportunity to do some hands on data manipulation. A particular feature of this course is that it provides these skills in a real world context of policy evaluation. It also focuses primarily not on experimental evaluation (Random Control Trials) but rather quasi-experimental methodologies that can be used where an experiment is not desirable or feasible.

This module introduces the statistical techniques of Exploratory and Confirmatory Factor Analyses. Exploratory Factor Analysis (EFA) is used to uncover the latent structure (dimensions) of a set of variables. It reduces the attribute space from a larger number of variables to a smaller number of factors. Confirmatory Factor Analysis (CFA) examines whether collected data correspond to a model of what the data are meant to measure. STATA will be introduced as a powerful tool to conduct confirmatory factor analysis. A brief introduction will be given to confirmatory factor analysis and structural equation modelling.

• Session 1: Exploratory Factor Analysis Introduction
• Session 2: Factor Analysis Applications
• Session 3: CFA and Path Analysis with STATA
• Session 4: Introduction to SEM and programming

In this module students will be introduced to meta-analysis, a powerful statistical technique allowing researchers to synthesize the available evidence for a given research question using standardized (comparable) effect sizes across studies. The sessions teach students how to compute treatment effects, how to compute effect sizes based on correlational studies, how to address questions such as what is the association of bullying victimization with depression? The module will be useful for students who seek to draw statistical conclusions in a standardized manner from literature reviews they are conducting.

Propensity score matching (PSM) is a technique that simulates an experimental study in an observational data set in order to estimate a causal effect. In an experimental study, subjects are randomly allocated to “treatment” and “control” groups; if the randomisation is done correctly, there should be no differences in the background characteristics of the treated and non-treated groups, so any differences in the outcome between the two groups may be attributed to a causal effect of the treatment. An observational survey, by contrast, will contain some people who have been subject to the “treatment” and some people who have not, but they will not have not been randomly allocated to those groups. The characteristics of people in the treatment and control groups may differ, so differences in the outcome cannot be attributed to the treatment. PSM attempts to mimic the experimental situation trial by creating two groups from the sample, whose background characteristics are virtually identical. People in the treatment group are “matched” with similar people in the control group. The difference between the treatment and control groups in this case should may therefore more plausibly be attributed to the treatment itself. PSM is widely applied in many disciplines, including sociology, criminology, economics, politics, and epidemiology. The module covers the basic theory of PSM, the steps in the implementation (e.g. variable choice for matching and types of matching algorithms), and assessment of matching quality. We will also work through practical exercises using Stata, in which students will learn how to apply the technique to the analysis of real data and how to interpret the results.

This module is for students who don’t plan to use quantitative methods in their own research, but who need to be able to read and understand published research using quantitative methods. You will learn how to interpret graphs, frequency tables and multivariate regression results, and to ask intelligent questions about sampling, methods and statistical inference. The module is aimed at complete beginners, with no prior knowledge of statistics or quantitative methods.

This intensive course on structural equation modelling will provide an introduction to SEM using the statistical software Stata. The aim of the course is to introduce structural equation modelling as an analytical framework and to familiarize participants with the applications of the technique in the social sciences.

The application of the structural equation modelling framework to a variety of social science research questions will be illustrated through examples of published papers. The examples used are drawn from recent papers as well as from publications from the early days of the technique; some use path analysis using cross-national data, others confirmatory factor analysis, and other still full structural models, to test particular hypotheses. Some example papers may be found below, though they should not be treated as the gold standard, rather as an illustration of the variety of approaches and reporting techniques within SEM.

• Duff, A., Boyle, E., Dunleavy, K., & Ferguson, J. (2004). The relationship between personality, approach to learning and academic performance. Personality and individual differences, 36(8), 1907-1920.
  
• Garnier, M., & Hout, M. (1976). Inequality of educational opportunity in France and the United States. Social Science Research, 5(3), 225-246.
  
• Helm, F., Müller-Kalthoff, H., Mukowski, R., & Möller, J. (2018). Teacher judgment accuracy regarding students' self-concepts: Affected by social and dimensional comparisons?. Learning and Instruction, 55, 1-12.
  
• Parker, P. D., Jerrim, J., Schoon, I., & Marsh, H. W. (2016). A multination study of socioeconomic inequality in expectations for progression to higher education: The role of between-school tracking and ability stratification. American Educational Research Journal, 53(1), 6-32.
  

Students will engage in a critique of such examples, with the aim of gaining a better understanding of the SEM framework, as well as its application to real-life data. To further facilitate this application focus, the theoretical introduction will be accompanied by practical examples based on real, publicly-available data.

This module introduces the time series techniques relevant to forecasting in social science research and computer implementation of the methods. Background in basic statistical theory and regression methods is assumed. Topics covered include time series regression, Vector Error Correction and Vector Autoregressive Models, Time-varying Volatility, and ARCH models. The study of applied work is emphasized in this non-specialist module. Topics include:

• Introduction to Time Series: Time series and cross-sectional data; Components of a time series, Forecasting methods overview; Measuring forecasting accuracy, Choosing a forecasting technique
• Time Series Regression; Modelling linear and nonlinear trend; Detecting autocorrelation; Modelling seasonal variation by using dummy variables
• Stationarity; Unit Root test; Cointegration
• Vector Error Correlation and Vector Autoregressive models; Impulse responses and variance decompositions
• Time-varying volatility and ARCH models; GARCH models

Longitudinal data analysis is a statistical method used to examine data collected from the same subjects or entities over multiple time points. This type of data analysis is particularly valuable for understanding how variables change over time and for investigating trends, patterns, and relationships within a dynamic context. For instance, how does children’s early home environment affect their future mathematical development?

Longitudinal data analysis holds several advantages, such as (1) understanding individual-level trajectories, enabling a deeper understanding of how different subjects respond to interventions or external factors over time, (2) supporting stronger causal inference by tracking changes before and after an intervention and (3) accounting for heterogeneity since it recognises that not all subjects respond uniformly to changes over time.

Over the course of this module, participants will learn how to work with longitudinal data. Through hands-on exercises and practical examples, participants will gain proficiency in data manipulation, visualisation, and advanced statistical techniques tailored specifically for longitudinal data. From understanding growth trajectories to uncovering causal relationships, this module will empower participants to navigate the complexities of longitudinal data with confidence. It is suitable for postgraduate students and researchers at any stages of their study and research. However, foundational Stata skills are required.

Panel data consists of repeated observations measured at multiple time points, collected from multiple individuals, entities, or subjects over a period of time. For instance, child A’s numeracy test score in Year 1, Year 2, Year 3 and Year 4. Country B’s GDP per capita in year 2020, 2021, 2022 and 2023. Panel data analysis, as a subset of longitudinal data analysis, is particularly useful for addressing research questions that try to understand how variables change over time and how individual units differ in their responses to changes. An example research question could be: how do children's numeracy scores vary across different socioeconomic backgrounds, and how have these disparities changed over the years? Panel data analysis holds several advantages, such as (1) increased statistical efficiency, (2) more effective at controlling for unobserved individual or entity-specific effects, and (3) more capable to study the dynamics of relationships over time.

Over the course of this module, participants will learn how to work with panel data. Through hands-on exercises and practical examples, participants will gain proficiency in data manipulation, visualisation, and advanced statistical techniques tailored specifically for panel data. It is suitable for postgraduate students and researchers at any stages of their study and research. However, foundational Stata skills are required.