Build a Practice You Can Change

By now, it would be easy to assemble a program. Pick a friend. Choose a day. Make a clear invitation. Repeat. Give the sequence a reassuring name and promise that loneliness will diminish if the reader follows it faithfully.

That would be a stronger promise than the evidence supports. It would also change the character of the friendship. Another person would become part of a treatment we had assigned ourselves without asking whether they wanted that role.

The practical argument of this book is more limited. Wanted friendship needs occasions in which people can take part. We can sometimes improve those occasions by making plans clearer, noticing costs and access, sharing work, and allowing the arrangement to change. These are things people can try. They are not a guaranteed cure for loneliness, and the studies of organized interventions do not turn them into one.

We should look closely at that evidence before deciding what to do next. There are promising results. There are also differences between measures, populations, comparisons, and follow-up periods that materially change what a result means.

What a trial of calls actually tested

Kahlon and colleagues randomized 240 Meals on Wheels clients in Texas to an empathy-focused telephone program or a no-call comparison during the COVID-19 pandemic. Calls ran for four weeks. On the three-item UCLA loneliness scale, which ranges from three to nine, the postassessment group difference favored the calls by 1.1 points, with a 95 percent confidence interval of 0.5 to 1.7. A different loneliness measure, the De Jong scale, did not show a statistically clear difference. Physical-health scores showed no clear benefit either.¹

This gives us a reason to take supportive contact seriously. It also gives us reasons to describe the result precisely. The participants were receiving a supported program in a particular setting. The endpoint came after four weeks. Thirteen participants were lost to follow-up in the intervention arm, compared with one in the comparison arm. The trial did not establish a minimum clinically meaningful difference for those scales.

The difference between the two loneliness measures is part of the result. We should not select the favorable scale and write as though every way of measuring loneliness produced the same answer. Nor can we turn the program’s schedule into a required frequency for friends.

A no-call comparison answers one question: what happens when this program is offered compared with not offering those calls? It does not tell us which element mattered most. Attention, expectations, conversation, the callers’ training, and the program’s organization may all contribute. A comparison with another active form of support can ask a different question.

Friendly attention can be the comparison

Choi and colleagues compared video-delivered behavioral activation with friendly video visits among 89 homebound older adults. Behavioral activation was a structured, coached approach to increasing meaningful activity; it was not simply an instruction to see friends more often. Both groups received contact. The initial program involved five weekly sessions.²

Bruce and colleagues later reported on the same trial at one year. Sixty-four participants were reinterviewed, and models used the available observations under assumptions about missing data. The reported advantage for behavioral activation had diminished but remained on modeled measures of loneliness, interaction, and satisfaction with support.³

These are two reports from one trial. Counting them as two independent replications would exaggerate the evidence. The follow-up is valuable because it asks whether a difference persists, but the missing participants introduce uncertainty. Statistical models can use the information available; they cannot recreate the experiences that were never observed.

An active comparison also makes the conclusion more specific. The question is whether a structured program adds something beyond friendly visits under the trial’s conditions. It does not establish that friendship itself is ineffective, and it does not mean an ordinary friend should start delivering behavioral treatment.

Those distinctions protect the role of friendship. A friend can help you take part in something you value, if that help is welcome. They do not need to become a coach, monitor adherence, or judge progress against a clinical outcome.

A larger trial still needs careful reading

The HEAL-HOA trial in Hong Kong enrolled 1,151 older adults who were lonely, living alone, below the poverty line, and digitally excluded. Trained older laypeople delivered telephone-based behavioral activation, mindfulness, or befriending. The initial intervention lasted four weeks, with two thirty-minute sessions each week. This was a selected population and an organized service, with befriending as an active comparison.⁴

The twelve-month report followed the same cohort. By then, 699 participants, or 60.7 percent, had dropped out, with different loss rates across the groups. The analysis reported small adjusted UCLA loneliness differences for behavioral activation and mindfulness compared with befriending, approximately seven-tenths of a point on the twenty-to-eighty scale. The standardized effects were small. Mindfulness did not show a clear advantage on the De Jong loneliness measure.⁵

A large starting sample does not make substantial follow-up loss unimportant. If people who remain differ from those who leave in ways related to outcomes, the model’s assumptions matter. Sensitivity analyses can examine alternative assumptions. They do not make the missing experiences visible.

The UCLA scale here is also different from the short version used in the Texas trial. A one-point difference on a six-point span and a one-point difference on a sixty-point span cannot be treated as the same unit of improvement. Even standardized effect sizes do not erase differences in population, program, and comparison.

The responsible conclusion is that some structured approaches show benefits in some settings, with uncertainty about their size, persistence, and wider application. That is a reason to improve and study support. It is not a reason to tell every reader to schedule two calls a week.

Reviews do not remove the need to inspect methods

A review can bring several studies together and reveal patterns that one trial cannot. It can also bring together studies that ask substantially different questions.

Shekelle and colleagues reviewed sixty studies, including thirty-six randomized trials, and found modest average reductions in loneliness for certain categories of interventions among community-living older adults. Group-based treatment and internet training had moderate-certainty evidence in their review. Many other categories had much weaker evidence.⁶

“Group-based treatment” should not be silently rewritten as “join any group.” The programs, participants, and support matter. Internet training is also different from simply increasing time online. A person may need practical help to use a tool before it expands their choices at all.

Blodgett and colleagues’ rapid review covered 95 studies and 101 interventions, including designs with and without control groups. That mixture requires care. Its controlled comparison for psychological interventions across sixteen studies was not statistically clear: the standardized difference was −0.64, with a 95 percent interval from −1.37 to +0.09.⁷

A reduction measured within one group before and after a program does not by itself isolate the program’s effect. People’s circumstances may change. An unusually difficult period may become less difficult. Participants may receive other help. Those who dislike the program may be less likely to provide the later measure. A comparison group can address some explanations, depending on how it is formed. Without one, we should resist treating every improvement as a treatment effect.

This is why a ranking of large pre-to-post changes can be misleading. It may rank differences in participants and methods along with differences in programs. It cannot automatically tell a person which intervention to choose, much less establish that one kind of friendship is best.

Different outcomes can lead to different conclusions

A review by C. Vazquez Alvarez and colleagues focused on social-connection interventions for people aged eighteen to twenty-four. It included six studies, mostly in higher education. Its pooled estimate for loneliness, drawn from three studies, was small and not statistically clear: the standardized mean difference was −0.10, with a 95 percent interval from −0.24 to +0.05.⁸

The interval includes zero, so the result does not establish a reliable reduction in loneliness. It also does not prove that no participant benefited. The review’s depression findings were a different outcome and should not be substituted for the loneliness result.

This is a recurring temptation in the literature and in everyday advice. We want a broad answer to whether connection helps. The studies measure narrower things: a particular questionnaire at a particular time, among particular people, after a particular intervention. Narrow answers can be useful if we allow them to remain narrow.

An umbrella review by Hansen and colleagues considered 29 systematic reviews of randomized-trial evidence. Most were rated low or critically low quality under its appraisal, and primary trials overlapped across reviews.⁹

Twenty-nine reviews do not mean twenty-nine independent confirmations. If the same small trial appears in several reviews, its repetition on the page does not increase the number of people studied. A careful synthesis asks how much distinct evidence exists, how strong it is, and whether the findings concern the same outcome.

This book’s source selection is a focused examination of relevant research, not a new comprehensive systematic review. Its analysis package makes two narrower calculations reproducible. Neither the quantity of references nor the presence of code should be mistaken for certainty beyond those tasks.

What the evidence changes about the advice

The studies give us several reasons to reject a universal program.

First, the people differ. Findings among selected homebound older adults or students cannot simply be assigned to every adult who wishes they saw their friends more often.

Second, the interventions differ. Trained callers, coaching, equipment, organized sessions, and active comparison conditions are part of what was tested. Removing those conditions while retaining the promised effect changes the claim.

Third, the outcomes differ. More messages, greater connection after a conversation, lower loneliness on one scale, and better friendship over years are related questions, not interchangeable results.

Finally, attendance and follow-up matter. The people who arrive, respond, or remain in a study may differ from those who do not. We saw that problem in the coffee-break experiment and again in longer intervention studies. A result among the people we can observe does not automatically describe everyone originally invited.

These points should lead to a more humane practice. If an arrangement does not help, the person has not failed to apply a universal law. We can reconsider the occasion, the desired outcome, the available support, or whether the relationship is wanted. We can also recognize when the problem exceeds what a friendship arrangement can address.

Persistent, distressing loneliness may warrant appropriate professional or community support, especially when accompanied by other difficulties. A friend can be part of that support while remaining a friend. The existence of treatment research does not create a duty for friends to deliver treatment.

Begin with one decision you can revise

You do not need a complete friendship plan before taking a useful step. Begin with a particular wish involving a particular person or occasion.

Perhaps you want to see an existing friend whose company you miss. Perhaps you want to return to an activity where conversation happened easily. Perhaps you want to stop being the only organizer. Those wishes call for different actions.

Name the obstacle as concretely as you can. If cost is the problem, an expensive dinner will not improve because it is scheduled earlier. If the problem is uncertainty about work, a booking deadline may make things harder. If the problem is hurt, another pleasant outing may avoid the conversation that is needed.

Choose the smallest honest proposal that addresses the actual obstacle. “Would you like lunch near your work?” “Could someone else organize next time?” “Would it be all right to talk about what happened?” The other person can answer, and that answer may change what you do.

Afterward, ask what you learned rather than assigning a score to the friend. Did the meeting happen? Was the arrangement manageable? Did both people seem able to choose? Did it offer the kind of contact you wanted? Was there a cost that one person absorbed silently?

A meeting can be enjoyable and too difficult to repeat. It can be easy to organize and emotionally disappointing. It can be awkward because the relationship is new, or because the people want different things. The point of looking back is to distinguish these possibilities, not to extract a number that pretends to explain them all.

You may keep the arrangement, change it, or stop. Either person can make that last choice without first completing a minimum trial. A practice is worth continuing when the people find it worthwhile and can sustain it, not merely because it has survived on a calendar.

Return to the dinner

The restaurant with the blue door remains in the group chat. Occasionally someone scrolls past it while looking for an address.

Nora and Dev have another lunch, then miss several weeks. Rosa and Eli manage the nearby outing on a Sunday when the morning decision works for both of them. Eli’s group call is pleasant for twenty minutes and disorderly after that. Dev loses the connection. Rosa leaves early. Nobody mistakes it for the long dinner they once imagined.

The friends do eventually find an evening when all four can meet. They choose somewhere cheaper. Before booking, Nora asks whether everyone can manage the price and the journey home. Dev confirms after his rota arrives. Rosa arranges care and says that she may still have to leave early.

There is no revelation at the table. Someone has been waiting to tell a story and discovers halfway through that two people have heard it. A subject becomes unexpectedly serious. The food takes too long. They disagree about dessert.

The evening matters partly because it happens at all. It also matters because nobody is required to pretend that arriving was easy, that the friendship is now repaired forever, or that the next dinner will arrange itself.

They leave with different journeys home and no shared free evening yet identified. Nora is no longer responsible for solving that fact alone. Someone will make another offer when they can. Someone else may decline it. There will be shorter meetings, messages, gaps, and changes.

Affection has not removed the conditions of their lives. They have become a little better at making those conditions visible and finding occasions that fit.

That is the practice worth keeping: take the wish to see someone seriously, make an offer they can understand, and leave enough freedom for both people to tell the truth about what comes next.