Two kinds of sudden gains in brief psychotherapy

Abstract

Objective

Sudden gains (SGs) are known to predispose to good outcome in psychotherapy, especially in brief treatment. However, some SGs may be illusory, in the sense that they arise from measurement error in the context of gradual change. We examined change before, during, and after SGs that were either true or illusory.

Method

In a sample of 1,867 university students treated with brief psychotherapy, we simulated session data as gradually changing score sequences, identified artifactual SGs therein, and utilized the simulated data to categorize actual participant SGs as either illusory or true.

Results

During treatment, participants with illusory SGs (N = 42) improved as much as did participants with true SGs (N = 67). Moreover, late in treatment participants with SGs and their matched controls improved at similar rates. True SGs were preceded by more distress and were larger than illusory SGs. Among participants with true SGs, very large SGs were more likely to reverse later in treatment.

Conclusions

Relatively small SGs may reflect measurement error. When brief psychotherapy patients deteriorate early in treatment but then suddenly experience substantial improvement, little further change is to be expected.

Keywords:

• psychotherapy
• process research
• sudden gains

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Notes

1 The Reliable Change Index (Jacobson & Truax, 1991) represents the minimum significant change between two scores. For p = .05, RCI = 1.96 x √2 x SEm = 1.96 x √2 x SD x (1 – r_tt)^½, where SD = the standard deviation of test scores, and r_tt = test-retest reliability. SEm is the standard error of measurement, reflecting the extent to which repeated test administrations vary around the examinee’s “true” score. In the current study, we calculated SD and r_tt (and hence SEm and RCI) from all participant data.

2 Lutz et al. (2007) proposed relaxing the third condition in order to identify SGs after the second treatment session or before the penultimate treatment session, a modification that has since been utilized in many SG studies. Pursuant to this proposal, we permitted the third criterion to be satisfied as follows: (a) with six consecutive sessions (three before the SG and three after the SG), or, in the absence of six consecutive sessions, with five consecutive sessions, either (b) two before the SG and three after the SG, or (c) three before the SG and two after the SG, subject to appropriate adjustment of the t-test critical value. Importantly, Vittengl et al. (2005) noted that the t-test is not an appropriate metric for this criterion, since t-tests assume independent observations, but consecutive session scores are unlikely to be independent. We retained this criterion to permit comparison with other studies.

3 A more intuitive strategy is to assign a participant SG to the true category if it “looks like” other participant SGs, or to assign it to the illusory category if it “looks like” artifactual SGs. In formal terms, assign an SG to the true category if p_ACT > p_ART, where p_ACT is the probability that the SG belongs to the distribution of all actual SGs and p_ART is defined as before. If p_ACT < p_ART, assign the SG to the illusory category. However, this strategy yielded unstable results in random subsample draws of participant SGs.

4 This result is roughly equivalent to the Cohen’s d value reflecting paired comparisons of multiply imputed first-session and final-session scores, d = .84, 95% CI = [.57–1.11], p < .001.

5 Our use of loglinear trajectories to model half of participant data had little influence on the summary data for artifactual SGs. Had we modeled all participant data using linear trajectories, artifactual SGs would have occurred in 6.6% (95% CI = [6.2% – 7.0%]) of simulated data, the median artifactual SG occurring in the fourth session. Reversals of artifactual SGs would have occurred 12.0% (95% CI = [11.1% – 12.8%]) of the time, the median reversal occurring in the seventh session. Notably, the use of linearly simulated data would have led to the identification of 43 SGs as illusory and 66 as true (cf. 42 illusory SGs and 67 true SGs, in the next section of the Results).

6 In the SG literature, missing data are not typically imputed to identify SGs, and while this analytic approach is undoubtedly correct, it is very likely to affect reported SG rates. Our 1.7% data loss resulted in a 6.7% decrease in the number of sequences in which SGs might have been identified. Monte Carlo data simulations indicate that if we had had a 10% data loss, there would have been 39% fewer sequences in which to identify SGs, and a 25% data loss would have decreased by 72% the number of such sequences. The effect of brief treatment is less dramatic but still significant. E.g., a 20% decrease in length of treatment from 15 to 12 sessions results in 30% fewer sequences in which SGs might be identified. A 33% reduction from 15 to 10 sessions results in 50% fewer sequences.