Case sampling for psychotherapy practice, theory, and policy guidance: Qualities and quantities

Abstract

Random sampling of cases is usually infeasible for psychotherapy research, so opportunistic and purposive sampling must be used instead. Such sampling does not justify generalizations from sample to population-distribution statistics, but does justify reporting what independent-variable value configurations are associated with what dependent-variable value configurations. This allows only the generalization that these associations occur at least that frequently in the population sampled from, which is enough for suggesting and testing some psychotherapy theories and informing some psychotherapy practice. Although psychotherapy practice is a longitudinal process, formal psychotherapy outcome research is so far most feasible and most widely done in the form of two-phase cross-sectional input-outcome studies. Thus, the analysis of sampling for psychotherapy research here will be in terms of the independent- and dependent-variable value configurations produced in such two-phase studies.

Il campionamento casuale dei casi è di solito impraticabile nella ricerca in psicoterapia, per questo si finisce per utilizzare un campionamento di comodo. Tale campionamento non permette di generalizzare i risultati dal campione alla popolazione, ma permette di identificare quali configurazioni della variabile indipendente sono associate con quali configurazioni della variabile dipendente. Ciò consente solo la generalizzazione secondo cui queste associazioni si verificano con almeno una determinata frequenza nella popolazione valutata, dato sufficiente per suggerire e testare alcune teorie psicoterapeutiche e portare nuove informazioni alle diverse pratiche psicoterapeutiche. Sebbene la psicoterapia sia un processo longitudinale, la ricerca di esito in psicoterapia è finora per lo più svolta nella forma di studi trasversali a due fasi ingresso-uscita. Pertanto, l'analisi del campionamento della ricerca in psicoterapia sarà svolta qui nei termini di configurazioni di valori di variabili indipendenti e dipendenti in studi a due fasi di questo tipo.

A amostragem aleatória de casos é geralmente inviável na investigação em psicoterapia, pelo que deverá ser utilizada como alternativa uma amostragem oportunista e intencional. Este tipo de amostragem não justifica generalizações a partir da amostra para as estatísticas de distribuição da população, mas justifica relatar que configurações de valor das variáveis independentes estão associadas com as configurações de valor das variáveis dependentes. Isto apenas permite generalizar que estas associações ocorrem com pelo menos alguma frequência na população que se retirou da amostra, o que é suficiente para sugerir e testar algumas teorias psicoterapêuticas, bem como para informar alguma prática psicoterapêutica. Apesar de a prática psicoterapêutica ser um processo longitudinal, a investigação formal em psicoterapia relativa aos resultados é muito mais viável e amplamente realizada sob a forma de estudos transversais de input-resultado, organizados em duas fases. Assim, a análise da amostragem para a investigação em psicoterapia será abordada neste estudo em função das configurações de valor das variáveis independentes e dependentes, conforme produzidas nos referidos estudos bifásicos.

Für die Psychotherapieforschung ist die zufällige Stichprobenziehung von Fällen normalerweise nicht durchführbar, weshalb eine opportunistische und zielgerichtete Stichprobenziehung stattdessen durchgeführt werden muss. Eine solche Stichprobenziehung rechtfertigt keine Generalisierung von Verteilungsstatistiken der Stichprobe auf die der Population, aber den Bericht darüber, welche Werte einer unabhängigen Variable mit welchen Werten einer abhängigen Variable assoziiert sind. Dies erlaubt nur die Generalisierung, dass diese Zusammenhänge mindestens so häufig in der Population auftreten, aus der die Stichprobe stammt, was ausreicht, um einige psychotherapeutische Theorien anzuregen und zu testen und um teilweise für die psychotherapeutische Praxis nützlich zu sein. Obwohl die psychotherapeutische Praxis ein Längsschnittprozess ist, ist die formale psychotherapeutische Ergebnisforschung in Form von zweiphasigen, querschnittlichen Input-Ergebnisstudien am praktikabelsten und am häufigsten durchgeführt. Deshalb wird die Analyse der Stichprobenziehung für die Psychotherapieforschung hier eine in Form der Wertezusammenstellungen unabhängiger und abhängiger Variablen, welche im Rahmen solcher zweiphasigen Studien gefunden wurden, sein.

心理治療研究通常不太可能用隨機抽樣選取個案，因此必需運用機會抽樣和立意抽樣。這樣的抽樣方法在統計上無法將抽樣樣本概括推估至母群分配，但可說明哪些自變項數值結構與哪些依變項數值結構有相關。這樣的結果僅能用來推論這些相關的連結在被抽樣的母群體中會經常出現，進而足以支持與測試一些心理治療理論及運用在心理治療實務。雖然心理治療實務是一個長期的歷程，但是兩階段、橫斷的輸入-效果研究仍是目前最可行也最被廣泛運用的正式心理治療成效研究法。因此，本文所介紹的心理治療研究的樣本分析是指這種兩階段研究中所產生的自變項、依變項數值結構型態。

Keywords:

• random sampling
• purposeful sampling
• opportunistic sampling
• case population
• hyperspace point population

Parole chiave::

• campionamento casuale
• campionamento mirato
• campionamento di comodo
• studi di casi

Schlüsselwörter::

• zufällige Stichprobenziehung
• zielgerichtete Stichprobenziehung
• opportunistische Stichprobenziehung
• Fallpopulation
• Hyperspace Punktpopulation

關鍵字：:

• 隨機抽樣
• 立意取樣
• 機會取樣
• 案例母群
• 超空間母群落點

Notes

1 This way of distinguishing between quality and quantity is meant to clarify what these words are to mean here and so hopefully avoid the morass of extant usages critically noted by, for example, Allwood (2012). Neuenschwander (2013) provides an illuminating history of these two concepts in science and mathematics and Wertz et al. (2011) of Psychology's efforts at qualitative research. Sale, Lohfeld, and Brazil (2002) argue that qualitative and quantitative research are incompatible alternative forms of research because they are about quite different things, something subjective and something objective, which differs from the distinction between them used here.

2 A case that is deviant and rare is a statistical outlier that may be such because the data on it are invalid and so properly must be censored in order to avoid biasing an average (see, e.g., Chatterjee & Hadi, 1986; Iglewicz & Hoaglin, 1993) or it may validly provide especially important novel information that certainly ought not be censored (e.g., Markou & Singh, 2003; Singh & Markou, 2004). Since each single case matters for psychotherapy practice and theory research purposes, valid data on each case are essential regardless of how deviant and rare or ordinary and common the case is, but most importantly so for the deviant and rare.

3 Some clinics, health maintenance organizations, preferred provider organizations, or administrative areas may be able to obtain rich enough census or random sample data on their own psychotherapy cases for their policy or even practice research purposes, but these organizations' cases cannot be considered sufficiently representative of other psychotherapy providers' cases until these too can be fully enough studied.

4 It is possible to back-map from effects of interest to their possible causes if sufficient data on prior events have already been gathered. This requires that the relevant causal event had somehow been covered in these data for every back-mapping from studied instances of this effect, perhaps as something made notable by some plausible theory. For example, given a case of PTSD, we explore for the traumatic stress that “must” have earlier occurred. Since a particular kind of effect may have alternative sufficient condition causes (rather than only a single necessary and sufficient one: Krause, 2010), back-mapping can be much more complicated than fore-mapping. So for the sake of simplicity, we shall deal here only with fore-mapping.

5 Data can describe a compound entity as a whole (e.g., a sum, number of constituents, or mean) or describe each of its several constituents (e.g., each addend). Ecological inference endeavors to create some description of individual entities from totals or averages of these (which remains disputed, e.g., Freedman, 2010, pp. 83–104; King, 1997/2013, 1999). Psychotherapy input, process, and outcome research is on individual cases, however, and so its results can be relevant for practitioners, theorists, or policy-makers without any recourse to ecological inference. The problem of such case-level research is, instead, how to securely generalize from available sample data, from some n cases, that cannot be relied upon to be accurately representative of whatever population distribution of N cases is sought to be generalized to. They cannot because for psychotherapy research, neither a census nor large enough if any random samples are presently feasible.

6 Study of any sample of cases may suggest additional or alternative dimensions in terms of which to study cases by providing grounds for further theory development (e.g., Gobo, 2004; Onwuegbuzie & Leech, 2007; Strauss & Corbin, 1998). In other words, by closely enough studying cases to explore richer or “thicker” descriptions of each case in any set of n cases than only descriptions in terms of the dimensions of some initial independent- and dependent-variable hyperspace, researchers can parallel the accumulation of a case occupancy structure over the M_XY points of this hyperspace with some progressive revision of the dimensions and dimensionality of their initial XY-hyperspace and so with some altering of their dimensional framework for the further development of grounded theory (see, e.g., Luborsky & Rubinstein, 1995). Adequate psychotherapy research is too costly and received psychotherapy theory and accounts of practice are still too sparse to limit the purpose of sampling merely to testing extant theories and practice-forms or for estimating case population frequency-distribution parameters for informing policy.

7 It is worth noting that a truly replicate performance of an experiment or survey, as logically required for proper meta-analysis, is the result of another kind of sampling: That from a population of experiments or surveys equivalently (a) designed and measured, (b) sampled from the same case population, and (c) analyzed. So such a replicate is generally only an hypothetical abstraction, something that rarely happens were it even possible to ensure or recognize as such. Thus, all the problems standing in the way of ensuring and recognizing the achievement of case-population-distribution representative samples in a descriptive hyperspace also stand in the way of achieving replications of experiments or surveys (see van der Veer, van IJzendoorn, & Valsiner, 1994) and so of doing meaningful meta-analyses, as Quenouille (1953, p. 228) had in effect noted.

8 Statistical significance testing has provided the main reason for maximizing n, but has become increasingly accepted to be logically flawed (e.g., Kline, 2013; Nickerson, 2000) although still having a powerful influence on what empirical studies get published (see, e.g., Bakker, van Dijk, & Wicherts, 2012; Hunter & Schmidt, 2004, pp. 493–509). However, to properly inform significance testing and confidence bounding, an n must be randomly drawn (e.g., Stuart et al., 1999, p. 1) unless the sample is considered to be the population of interest and so n = N and significance testing is irrelevant. Furthermore, what is most important substantively for practice and theory purposes is how nearly n cases fully span and saturate their XY-hyperspace just like the N cases they are drawn from does (i.e., how nearly m_X represents M_X). For socially responsible policy purposes, how nearly the n represent the shape and location of these N cases' distribution in this space is obviously also important.