Multiple observations of one case

In most circumstances an EvalC3 data set will describe multiple attributes of multiple cases, where each case is represented by one row.

However in some circumstances there may only be one case that is of interest , but it may be that multiple observations can be made of the attributes of this case over a period of time. For example, a single project that varies its approach over a  period of time. Or a single family, whose welfare and wider circumstances vary over time. In both of these situations each row in an EvalC3 data set can represent a set of observations made at a given period of time.

Where there are many observations made over an extended period of time sampling issues may need to be considered. For example, whether to include all observations, or only the most recent, or only a moving fraction. The reason for the latter is that a given predictive model may only apply for a particular period of time.  This may be for good or bad reasons. See the wikipedia entry on  Goodhart’s Rule and Campbell’s Law

Postscript 1: There is a body of literature on single case research designs, which the above is an example of. Here is a quote re these designs:

“Single-case research designs (also referred to as “single subject designs”, “single-case experimental designs”, and “n-of-1 trials”; henceforth, SCRDs) have been used to assess intervention effects for many decades (Barlow & Hayes, 1979; Herson & Barlow, 1976). In contrast to experimental designs that involve comparing average outcomes across groups of individuals in different treatment conditions, SCRDs involve introducing an intervention to an individual case or cases and measuring changes in outcomes over time. Some types of SCRDs also involve removing and then re-introducing the intervention, providing further tests of the functional relationship between the intervention and the outcome. SCRDs are critically important for understanding the effectiveness of interventions for individuals with low incidence disabilities (e.g., physical disabilities, autism spectrum disorders), given the inherent difficulties in obtaining sufficient samples sizes for between-group experimental designs with such populations. As a result, SCRDs comprise a large part of the evidence base in certain areas within fields such as special education and school psychology. The results of SCRDs can under some circumstances provide a strong basis for understanding the causal effects of interventions (Gast & Ledford, 2014). They have the added advantage of providing information about intervention effects at the level of individual cases, whereas between-group experimental designs are informative only about average effects. Thus, the results of SCRDs are relevant for informing clinical and public policy decisions, and should be considered for inclusion in systematic reviews and meta-analyses that aim to synthesize the existing evidence about intervention effects (Council for Exceptional Children Working Group, 2014; Kratochwill et al., 2013).”

Valentine, J. C., Tanner‐Smith, E. E., Pustejovsky, J. E., & Lau, T. S. (2016). Between-case standardized mean difference effect sizes for single-case designs: A primer and tutorial using the scdhlm web application. Campbell Systematic Reviews, 12(1), 1–31.

Postscript 2: I have just read a 2020 paper by Sofia Pagliarin and and Lasse Gerrits pointing out that it is possible to have, and analyse, a set of cases that combine multiple different cases A-Z, along with multiple versions of some cases that are described at different points in time: At1, A t2, etc

Postscript 3: Configurations i.e. characteristics of a case may change over time. If we treat different aggregated time periods (A-J =0 versus K-Z = 1) as the “outcome” to be predicted, we could develop predictive models which summarise the core features (=a model) of each aggregated time period. An aggregated time period could be 1960-1970, or a presidential term, or an agricultural season.

%d bloggers like this: