Temporary deteriorations within the therapeutic relationship are a distinctive feature of change over the course of the processes, which was also noted in our sample. The general question we raised is: what crises are useful and which ones will harm the process? To create a systematic access, we proposed to distinguish three characteristics of therapeutic crises: number, magnitude and length.
The number of identified crises varies significantly, depending on the criterion used for measurement. Stiles et al.  pointed out that their rupture-repair criterion was developed ad hoc and should be regarded as preliminary; they indicated that application to other samples remained necessary. Strauss et al.  remarked that their method of quantifying alliance ruptures "is most similar to that of Stiles et al. (2004)" (p. 343). However, our testing yielded differing results, depending on the criterion that was used. Thus, the comparability of the results of studies which used different criteria seems to be limited in terms of the frequency of the ruptures and the ascertained number of courses with at least one rupture.
Stiles et al. and Strauss et al. used different thresholds for identifying ruptures. In doing so, they implicitly predefined at which magnitude a crisis is identified as one. We propose to use both, the stricter (2 sd) and the less strict threshold (1 sd) in order to ensure the comparability of the results of different studies and for the purpose of analyzing the extent of deteriorations in the therapeutic relationship in connection with the outcome.
We noticed weak points in the criteria definition of Stiles et al. and Strauss et al. and suggested several modifications. We propose to combine the intraindividual with the interindividual variability and to favor the stricter of the two values in each case, in order to avoid overrating minor fluctuations occurring in relatively stable courses or declines occurring in highly fluctuating courses which are intraindividually less significant. We propose to disregard the exclusion criteria, especially in connection with outcome studies, as this approach seems to contain a circular argument. We propose to classify cases in which two or more consecutive decreases fall below the threshold as one rupture-repair episode only. We recommend adjusting the repair value to the rupture value as, in our view, the crisis within the therapeutic relationship can only be seen as overcome if the previous level is reached again.
In comparison with the criteria of Stiles et al. and Strauss et al., more ruptures were identified with the crisis-repair criterion proposed by us. Using our criterion, it becomes possible to identify crises developing in small steps over a number of sessions; in this way, we also captured gradual downward trends.
We differentiated five crisis-repair subtypes. The most frequent pattern was the simple V-shape ("jump in - jump out", 40.6%). The second most common pattern was a decline over more than one session including a jump with a sudden repair ("slide in including one or more jumps - jump out", 31.9%). It was striking that the episodes in more than half the cases showed more complex progress than a simple high-low-high pattern. It appeared that both the crises and the repairs may extend over several sessions. The length of the crises in the complete sample averaged 1.8 and the length of the repairs 1.2 sessions. The longest downward trend comprised a period of 6 sessions. In this light, we can assume that crisis-repair episodes exist at different temporal levels in the process. Like other authors proved, ruptures can be found within a single session that are repaired before the end of the same session [8–11]. Besides this, there are ruptures that are repaired in the following session (which results in a high-low-high pattern) and there are ruptures that extend over several sessions. Stiles et al. remarked that their criteria "were crude, justified by making use of ratings that were gathered only once for each session". They further stated that "moment-by-moment ratings of the session process are potentially much more sensitive, albeit more laborious" (p. 91). Alternatively, we can assume that the length of the episodes is related to specific characteristics of the therapeutic process. For example, it can be hypothesized that there is a relation to the extent of the entanglements of the therapists with a pathological hostile interaction pattern of the patient. Prolonged crises may potentially occur in more disturbed patients with severe interpersonal problems. Less severe interpersonal problems may lead to more subtle crises that may be resolved within the same or at least within the next session. It is also conceivable that the length of the episodes is negatively related to therapeutic qualities such as the professional experience.
As a result of the present observations, it would be interesting for future analyses to examine the potential influence of the mentioned variables on the length of the crisis-repair episodes and the effect of the length of the episodes on the outcome. For this purpose it is necessary to define in which session a crisis begins and in which session it should be considered repaired. Moreover, it seems to be useful to exactly define an episode to analyze the sessions in detail in order to reveal the mechanisms behind the crisis-repairs in the course of the therapeutic relationship. To create a basis for this kind of objective, it makes sense to examine the criteria that determine what a rupture or what a rupture-repair sequence is in depth.
At least one jump was seen in 86.4% of the declines. Thus, discontinuous sudden declines and increases in the trajectories of the therapeutic relationship are a central element of therapeutic progress. However, it also must be stated that we found gradual downward trends (isolated or in combination with a jump) in more than half the cases. When looking at the repairs, we can state that the gradual trend played a less important role. We found an isolated jump in 84% of the cases which lets us conclude that a breach in the therapeutic relationship can be dissolved completely immediately after focusing in most cases.
We have shown that the choice of a criterion influences the findings. We can not draw any unambiguous conclusions about which is the best methodology to use but we may provide support to researchers in making an informed and thoughtful decision about how to proceed, depending on the context. Strictly speaking, it must be decided beforehand whether leaps (discontinuities) or crises of varying length can be assumed to be the crucial mechanism of change. If we theoretically view the leap to be crucial, the modified criterion of Strauss et al. could be used. If we theoretically focus on the importance of extreme lows, the modified criterion of Stiles et al. appears to be useful. The advantage of using the criterion proposed by us is that ruptures with different patterns and lengths (including gradual trends) are included. This means we regard all crises, both gradual and sudden deteriorations, in the experienced therapeutic relationship as important mechanisms of change without limiting our scope to single discontinuities, i.e. sudden deteriorations. This provides a wider range for different assumptions.
We would like to point out that it is useful to ascertain variables continuously throughout the processes. The results may be distorted if only selected sessions are examined as jumps between the measurements can occur. The study of Strauss et al. was conducted with larger intervals between the measurements. A part of the ruptures found in the study of Strauss et al. may have been developed gradually over several sessions or the number of ruptures may be even higher than documented.
It also should be mentioned that the number of identified ruptures might depend on the measuring instrument. Stiles et al. used the Agnew Relationship Measure , while Strauss et al. used the California Psychotherapy Alliance Scale . It is possible that certain different aspects of the therapeutic relationship are detected by different instruments. It might be that the Intrex [24, 25] is a particularly sensitive instrument for showing fluctuations in the therapeutic relationship as it measures patterns of the therapeutic relationship (the interaction between therapist and patient) directly and in a relatively differentiated manner.
In our present state of knowledge, many questions still remain open. Further analyses on this topic are required. This may be unsatisfactory for the time being, as we can not provide hard data at this first stage. Nevertheless, our approach can contribute to pave the way for further steps in this kind of research thoroughly and with a solid grounding. A phenomenological discussion of the conception of crises is useful in order to distinguish specific aspects and to specifically link these characteristics to the treatment outcome in future studies. This kind of systematic research offers the advantage that conclusions about both, positive and negative correlations can be drawn in the future. Empirical studies of the therapeutic relationship and of the factors influencing that relationship are essential to clinical practice as change always emerges from the context of the patient-therapist relationship independently of the applied therapeutic technique. The findings of the study give an insight into basic mechanisms of change within the therapeutic relationship. After all, the methodological concerns we address might be adaptable to different research areas where the analysis of fluctuations in a variable of interest over time is relevant.