skip to content

Department of Psychology

Six classes in Lent Term of 1.5+ hours’ duration. Postgraduates in all years are welcome and on special request a few from other related departments can typically be accommodated. Some of the statistically most knowledgeable students have found it worthwhile to re-attend for specific topics..

All sessions will take place in Lent 2021.

Detailed requirements for postgraduate projects in respect of methods are diverse. (‘Statistics’ is too narrow a term here, although ‘strategic uses of statistics’ comes close to what is coverd.)  After this taster Professor Haggard is available through the year for (unpaid!!) consultancy to those who have attended it.

The so-called Replication Crisis (RC) shows that decades of studies deemed publishable have done things wrongly. A more profound reading shows they have largely been attempting to do the wrong thing (to claim specious factoidality -- finding one example of the possible existence of some effect that it is not outrageous  to claim so will survive in the publication lottery. We now consider it is more important to go directly to specifying the effect its nature and magnitude. Rather than assuming that after the suggestion of some association or link having been ‘discovered’ by the author, some different type of scientist is going to come along and determine its magnitude and range of generalisation, to go directly for these scientific goals.).

This course attempts to transfer understanding of statistical concepts, so that the self-deceptions of RC do not occur in the future. It emphases the scientific goals to be achieved by studies: generalisability, stability, power by error-reduction, parsimony and the explanation of worthwhile proportions of the variance (effect size). It includes some initial familiarisation with the necessary statistical techniques for data structures more complex than those met in typical undergraduate experimental projects with their simple differences, correlations and ANOVA designs. The discussions are rooted in issues that make a difference: what are the specific (psychological AND statistical) advantages AND disadvantages of between- versus within-participant designs? What is the p-value really, and what should it be used for, if anything? For what reasons should you test for interaction, how many such tests can you support, and what ought you to do you next, especially if the interaction result is marginal as to ‘significance’? If you have a difference carried by a higher level of sampling unit such as school or hospitals, then when and how do you need  to test for generality across these (MLM) rather than for some effect existing when totalled across participants within the data?

Well, the cover-all answer to all such questions is that it depends on the exact expression study of the particular research question for the  not just  the general research question or prediction. The course teaches that the generation of particular useful answers depends on having a set of methodological principles (eg that you must always show you have made use, to the extent feasible, of all relevant data obtained;; and the implementation of an answer depends on sufficient understanding of statistics to design a study and design a good analysis of it.  In 5 weekly sessions we meet the various powerful extensions of the General Linear Model (that is basically, of ANOVA) such as Multiple Regression, Factor Analysis and Multi-level modelling; also logistic regression and computationally intensive non-parametric methods and brief introduction to  Structural Equation modelling (SEM). The sixth class is a more practically oriented workshop in SEM.

Multiversing is emphasised: the need for generalisation reasons, met especially when a result is marginal or data are missing to do conceptually similar analyses with more than one method, bridging statistical, metrical and other differences. It is one of the preventatives and treatments for the symptom of non-replication in

Reading. PDFs on particular methodological points are supplied, but no post-graduate in psychology should proceed without having read in their first few months all three of:

P. Abelson. Statistics as Principled Argument, 1994. This a wise and humorous book on how to work towards as much general methodology as you ought to have in your psychology

G. Cumming. Introduction to the New Statistics, 2012. (This is a more stripped down presentation of how  statistics in psychology and many other empirical sciences ought to have been taught all along.  Beware, there is also a more recent book of similar title by Geoff Cumming which is OK, but more an undergraduate workbook.)

S Ritchie. Science fictions. Bodley Head, London 2020. An up-to date take on the replication crisis setting out the moral issues in the replication crisis.