Small Effects at National Scale: The Prevention Paradox and the Helping Relationship

Modest effects, large populations

A story told in many versions has a child on a beach after a storm, throwing stranded starfish back into the sea. An adult says that the beach runs for miles and the starfish number thousands, so the effort cannot matter. The child throws another and answers that it matters to that one. The story is usually traced to "The Star Thrower", an essay by the American anthropologist Loren Eiseley (1969), in which a narrator meets a man returning starfish to the sea. The child and the closing answer belong to later retellings, which vary.

The story makes a case for the value of help to each person reached. The question in this essay is a different one: whether modest effects on very large numbers of people can matter more in total than large effects on a few.

In psychotherapy, the strength of the working alliance, the collaborative bond between therapist and client, is linked to outcome by a correlation of about r = .278 across 295 independent studies, which by simple arithmetic accounts for roughly 8% of the variation in outcomes (Flückiger et al., 2018). Stronger social relationships are associated with 50% higher odds of survival (an odds ratio of 1.50) across 148 studies (Holt-Lunstad et al., 2010). YFL's essay Does the Helping Relationship Heal? What the Evidence Shows sets out both findings. Effects of this size can be read as modest. The question is whether modest effects per person are the right measure of importance when the thing in question is spread across a whole service, such as a children's social care department, or a whole country.

Rose's principle

Geoffrey Rose (1926-1993) was a British physician and epidemiologist (a specialist in how disease is distributed in populations) who held the chair of epidemiology at the London School of Hygiene and Tropical Medicine from 1977 (Royal College of Physicians, Lives of the Fellows). In a 1981 paper on cardiovascular disease, and in a 1985 paper titled Sick Individuals and Sick Populations, he separated two questions: what determines which individuals become cases, and what determines how many cases a population has. The two can have different answers. The 1985 paper's abstract describes two matching strategies, a high-risk approach that protects susceptible individuals and a population approach that seeks to control the causes of incidence, meaning the number of new cases. Rose reported that the two are not usually in competition, and that the prior concern should be to discover and control the causes of incidence (Rose, 1985).

A widely cited consequence is that a large number of people at small risk can give rise to more cases than a small number at high risk. A related consequence, described in a review of Rose's 1992 book, is that a preventive measure bringing large benefit to the community may offer little to each participating individual (Fowler, 1992). Rose's own work on heart disease is the evidence base for the principle.

What the arithmetic implies

The numbers below are invented for illustration. They are not data.

Setting. Suppose 100,000 families, of which 1,000 are judged high risk and 99,000 are not. Suppose that in a given year a bad outcome occurs in 20% of the high-risk families and 2% of the others.

Cases. The high-risk group produces 200 cases. The larger group produces 1,980. Most cases come from families that nobody would have flagged.

Two strategies. Intensive help for the 1,000 high-risk families, however effective, can address at most 200 cases. A small relative improvement across all 100,000, say a one-tenth reduction in the 2% rate, would prevent about 198 cases among the large group, and each participating family would see little difference.

The example shows the structure of the argument, not its truth for children's services. Real risk is not divided so neatly, and risk prediction in this field is imperfect. Applied to helping relationships, relational quality would take the part of the risk factor: a characteristic spread across the whole population, where a small shift in the average could change the number of bad outcomes.

Where the principle has support

Rose's reasoning has been revisited since. A 2006 review cited an evaluation of 122,458 patients in 14 international trials in which one in five men with coronary disease had none of four conventional risk factors, an observation consistent with disease arising among people not classed as high risk (Doyle et al., 2006). The same review and later commentary also record that population approaches often yield small benefit to each individual, which creates difficulty for the balance of risk and benefit.

Later literature has also raised challenges: that better identification of high-risk people may reduce the need for population-wide strategies, and that some population strategies may widen inequalities. Both challenges are set out at the start of a 2010 paper in the International Journal of Epidemiology that argues population prevention need not worsen inequalities (McLaren et al., 2010). That paper is itself a secondary source for them, and the original papers have still to be read. A related paper argues that screening and treating high-risk individuals typically widens inequalities, while whole-population measures can reduce them (Capewell & Graham, 2010).

The step that has not been tested

Applying the principle to helping relationships needs three further steps. None was found tested in the sources gathered for this essay.

Distribution. Whether poor relational quality in children's services is spread widely, or concentrated in a few services or workers, is unknown. In a study of 331 patients seen by 80 therapists, the alliance varied between therapists, and that variation predicted outcome, whereas variation between patients was unrelated to outcome (Baldwin et al., 2007). This suggests relational quality is not uniform among therapists.

Shifting the whole distribution. Whether training, caseloads or supervision can shift relational quality across a service is a separate question. A randomised trial of training in motivational interviewing, a counselling method, for child protection social workers has been reported. In one local authority, the training package produced a difference in rated skills that the authors describe as not substantial, and no statistically significant difference in other outcome measures, including parental engagement (Forrester et al., 2018). Evaluations of Family Safeguarding Hertfordshire, a reform of children's services across one English county, describe a whole-system change (Forrester et al., 2017; Rodger et al., 2020). YFL's essay Alliance Evidence in Mandated Work: What Transfers to Child Protection sets out the trial and the evaluations. Neither isolates the relationship.

Net benefit. A small benefit per family can be wiped out by a small harm. The Cambridge-Somerville Youth Study (a crime prevention programme for boys in Massachusetts, United States) and Scared Straight programmes (which bring young people into contact with prison life) are documented cases where programmes judged beneficial by those who ran them were reported to do worse than no programme (McCord, 2003; Petrosino et al., 2013). YFL's essay Cures Ahead of the Evidence: When Interventions Judged Beneficial Were Tested sets out both cases, including a 72-year follow-up of the Cambridge-Somerville participants in which the mortality difference seen in middle age was not observed in old age. A population-wide relational programme would need the same test.

Hypothesis, not finding

The following is a hypothesis. It follows from Rose's reasoning, but no direct evidence for it in children's services was found in the sources gathered for this essay: because relational quality is possibly distributed across many families and workers, modest improvements in its average might prevent more harm in total than intensive work with the few families judged highest risk. Testing it would need a measure of relational quality, a comparison group, and a record of harm as well as benefit. YFL's essays Sure Start: What Was Promised, What Was Measured, and When It Was Known and Examining the Evaluators: Why Are They Listened To? discuss what a fair test of this kind involves.

Rose's principle gives a reason not to dismiss moderate effects. It does not supply evidence that those effects operate in children's services, and it carries a warning that small harms spread widely also count. YFL offers the reasoning for readers to weigh, without instruction.

---

References

Baldwin, S. A., Wampold, B. E., & Imel, Z. E. (2007). Untangling the alliance-outcome correlation: Exploring the relative importance of therapist and patient variability in the alliance. Journal of Consulting and Clinical Psychology, 75(6), 842-852.

Capewell, S., & Graham, H. (2010). Will cardiovascular disease prevention widen health inequalities? PLoS Medicine, 7(8), e1000320.

Doyle, Y. G., Furey, A., & Flowers, J. (2006). Sick individuals and sick populations: 20 years later. Journal of Epidemiology and Community Health, 60(5), 396-398.

Eiseley, L. (1969). The star thrower. In The Unexpected Universe. Harcourt, Brace & World.

Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018). The alliance in adult psychotherapy: A meta-analytic synthesis. Psychotherapy, 55(4), 316-340.

Forrester, D., Lynch, A., Bostock, L., Newlands, F., Preston, B., & Cary, A. (2017). Family Safeguarding Hertfordshire: Evaluation report. Department for Education.

Forrester, D., Westlake, D., Killian, M., Antonopoulou, V., McCann, M., Thurnham, A., Thomas, R., Waits, C., Whittaker, C., & Hutchison, D. (2018). A randomized controlled trial of training in motivational interviewing for child protection. Children and Youth Services Review, 88, 180-190.

Fowler, G. (1992). Review of The strategy of preventive medicine by Geoffrey Rose. British Journal of General Practice, 42(364), 495.

Holt-Lunstad, J., Smith, T. B., & Layton, J. B. (2010). Social relationships and mortality risk: A meta-analytic review. PLoS Medicine, 7(7), e1000316.

McCord, J. (2003). Cures that harm: Unanticipated outcomes of crime prevention programs. Annals of the American Academy of Political and Social Science, 587, 16-30.

McLaren, L., McIntyre, L., & Kirkpatrick, S. (2010). Rose's population strategy of prevention need not increase social inequalities in health. International Journal of Epidemiology, 39(2), 372-377.

Petrosino, A., Turpin-Petrosino, C., Hollis-Peel, M. E., & Lavenberg, J. G. (2013). 'Scared Straight' and other juvenile awareness programs for preventing juvenile delinquency. Cochrane Database of Systematic Reviews, 2013(4), CD002796.

Rodger, J., Allan, T., & Elliott, S. (2020). Family Safeguarding: Evaluation report. York Consulting LLP for the Department for Education.

Rose, G. (1981). Strategy of prevention: Lessons from cardiovascular disease. British Medical Journal, 282(6279), 1847-1851.

Rose, G. (1985). Sick individuals and sick populations. International Journal of Epidemiology, 14(1), 32-38.

Rose, G. (1992). The strategy of preventive medicine. Oxford University Press.

Royal College of Physicians. Rose, Geoffrey Arthur. In Lives of the Fellows (Munk's Roll), Vol. IX, p. 451.