Is Low Therapist Empathy Toxic?

Abstract

One of the largest determinants of client outcomes is the counselor who provides treatment. Therapists often vary widely in effectiveness, even when delivering standardized manual-guided treatment. In particular, the therapeutic skill of accurate empathy originally described by Carl Rogers has been found to account for a meaningful proportion of variance in therapeutic alliance and in addiction treatment outcomes. High-empathy counselors appear to have higher success rates regardless of theoretical orientation. Low-empathy and confrontational counseling, in contrast, has been associated with higher drop-out and relapse rates, weaker therapeutic alliance and less client change. The authors propose emphasis on empathic listening skills as an evidence-based practice in the hiring and training of counselors to improve outcomes and prevent harm in addiction treatment.

In discussions regarding the merits of evidence-based addiction treatment, prominent attention has focused on the effect of therapist variables on behavior change (Imel, Wampold, & Miller, 2008; Morgenstern & McKay, 2007). Indeed, it appears that one of the strongest determinants of clients’ outcomes in addiction treatment in particular is the counselor to whom they happen to be assigned (Luborsky, McLellan, Diguer, Woody, & Seligman, 1997; Luborsky, McLellan, Woody, O’Brien, & Auerbach, 1985; Kraus, Castonguay, Boswell, Nordberg & Hayes, 2011; McLellan, Woody, Luborsky, & Goehl, 1988; Miller, Taylor, & West, 1980; Valle, 1981). Research consistently shows that differences among therapists account for between 5 and 12% of the variance in a variety of client outcomes, including substance use (Elliot et al., 2011) and that a better relationship between the client and therapist is associated with higher levels of treatment engagement and retention in substance abuse programs (Meier, Barrowclough & Donmall, 2005). Empirically-based substance abuse interventions such as cognitive-behavioral treatment, twelve-step facilitation and motivational interviewing rely at least in part on the interpersonal skills of the provider for their impact, yet little research exists concerning which skills or attributes contribute to variation in the quality of the therapeutic interaction.

Psychotherapy research generally has suggested that therapist differences may be attributable in part to outlier counselors with unusually adverse or particularly good client outcomes (Okiishi, Lambert, Nielsen, & Ogles, 2003; Shapiro, Firth-Cozens, & Stiles, 1989; Wampold & Bolt, 2006). In the area of substance abuse treatment more particularly, at least four studies have reported therapists with unusually poor client outcomes. In a multisite clinical trial (Project MATCH Research Group, 1998), therapist differences were no longer significant after removing one or two outliers in each treatment condition whose clients showed particularly poor drinking outcomes. In a naturalistic experiment following the resignation of two drug counselors, McLellan and colleagues (1988) randomly reassigned their 62 cases to four other counselors. This allowed them to observe differences in outcomes for these reassigned clients as a function of the new counselor to whom they had been assigned. Relative to their functioning at the time of reassignment, the clients of three of these counselors showed varying degrees of improvement on all measures, but a fourth counselor’s caseload showed increased rates of drug-positive urines, methadone dosage, and unemployment, and no reduction in arrests. In another clinical trial reported by this same group, one of three therapists providing supportive-expressive therapy had clients whose drug use on average increased during treatment, in contrast to significant improvement of cases assigned to two other therapists delivering the same manual-guided treatment (Luborsky et al., 1985). Finally, among clients randomly assigned to nine counselors providing manual-guided behavioral self-control training, the rates of within-caseload adverse outcomes ranged from zero to 75% (Miller, et al., 1980).

What may account for such differences in efficacy among therapists treating substance use disorders? Reference is often made to common or nonspecific factors that influence outcome regardless of the particular theoretical orientation of a therapist (Hubble, Duncan, & Miller, 1999; Wampold, 2001). Evidence points in particular to therapists’ interpersonal skills as a predictor of outcome (Anderson, Ogles, Patterson, Lambert, & Vermeersch, 2009; Valle, 1981). Just how common such skills are among counselors is unclear, but it is evident at least that they are not universal. If these “nonspecific” skills do indeed exert such a strong influence on outcome, then it is important to specify, screen for, and teach them, rather than viewing them as nuisance variables to be controlled (Morgenstern & McKay, 2007; Norcross & Wampold, 2011).

A prominent candidate among the therapist skills described as “nonspecific” is that of therapist empathy. The term empathy has many meanings, and it can be difficult to differentiate it from other relationship elements such as acceptance and warmth (Elliot et al., 2011). We know that clinician and client assessments of the quality of empathy in the same session can differ, and that they may both diverge from the ratings of independent observers. Usually, it is the client’s rating of their counselor’s empathy that is most predictive of outcomes, with those of independent observers being less powerful but still useful (while those of therapists about their own empathy have no relationship to outcome). This suggests that raters may be capturing different dimensions of the construct of empathy or even measuring different constructs entirely.

We are focusing here on what Carl Rogers and his students described as accurate empathy (Rogers, 1959; Truax & Carkhuff, 1967). This is a specific therapeutic skill that includes a commitment to understanding the client’s personal frame of reference and the ability to convey this heard meaning back to the client via reflective listening. This perspective-taking process encompasses the accurate understanding of both cognitive and emotional aspects of the client’s experience as well as attunement to the unfolding experience of a client during a treatment session (Elliot et al., 2011). Viewing empathy in this way has the advantage of focusing on specific clinician behaviors (i.e. reflective listening) that can be measured during treatment sessions, but it may underestimate the client’s contribution to it. Clinicians are clearly different from each other in the amount of empathy they typically convey in treatment sessions, but there are also differences within clinicians depending on client characteristics (Carkhuff & Alexik, 1967). It may be easier to be empathic with some clients than with others. Nevertheless, focusing on the clinician’s contribution to the empathic process is important because it is clinicians, and not clients, who are tasked with improving the quality of therapeutic interactions.

Counselor empathy is often described as a component of therapeutic alliance (Anderson et al., 2009; Baldwin, Wampold & Imel, 2007;Horvath, De Re, Flückiger & Symonds, 2011) and even a necessary ingredient (Boardman, Catley, Grobe, Little, & Ahluwalia, 2006; Feller & Cottone, 2003; Hoaas, Lindholm, Berge, & Hagen, 2011; Meissner, 1996; Rogers, 1959). Baldwin et al. (2007) have demonstrated that the robust relationship between therapeutic alliance and client outcomes is driven by therapist, and not client, variability in the alliance construct. They note that improved treatment outcomes could be expected if therapists were trained to develop and maintain strong alliances, including the capacity for genuineness, empathy and unconditional positive regard. Similarly, the American Psychological Association Task Force on Evidence-Based Therapy Relationships (Norcross & Wampold, 2011) has designated empathy as an evidence-based element of the therapuetic relationship and has recommended that training programs implement competence-based criteria for educating practioners in relationship elements.

In addiction treatment more particularly, a review of brief interventions for problem drinking similarly found that an empathic counseling style was a common component of effective interventions (Bien, Miller, & Tonigan, 1993). In addition, five studies have directly examined the specific relationship between client substance use outcomes and the empathic skill level of individual therapists. Miller, Taylor and West (1980) randomly assigned clients to nine counselors who all delivered manual-guided behavior therapy for problem drinking. The therapists were also trained in reflective listening, and three supervisors observed their sessions via one-way mirrors, rating their skillfulness in accurate empathy using a scale developed for this purpose by Truax and Carkhuff (1967). After the treatment phase but before examining outcome data, the supervisors independently rank-ordered the nine counselors on their level of skill in accurate empathy, with good inter-rater agreement. At six-month follow-up, the percentage of each counselor’s cases with positive outcomes was computed. The correlation between therapists’ empathy rank and a behavioral outcome measure – number of standard drinks consumed per week – was .r = 82, accounting for two-thirds of the variance in client drinking at 6 months. At 12 months the relationship was r = .72, accounting for half of the outcome variance. Even two years after treatment, 25% of outcome variance (r = .52) was still accounted for by therapist empathy (Miller & Baca, 1983).

This study (Miller et al., 1980) also included a comparison group randomly assigned to be sent home with a self-help manual (Miller & Muñoz, 2005) and encouraged to follow its instructions, then return in 3 months for follow-up. No additional treatment was provided for this group, and the percentage of positive outcomes for this “bibliotherapy” group (60%) was similar to the average success rate for counselor-facilitated treatment using the same behavioral approach (65%). This could lead to the conclusion that “therapists are no different from self-help,” which on average was true, but five of the therapists – primarily those with high empathic skills – had higher success rates than the self-help control group. One showed the same outcome rate, and in the case of three lower-empathy therapists it seems the client might have been better off going home with a good book.

Valle (1981) similarly rated the Rogerian interpersonal functioning of therapists in an alcohol treatment program. Empathy was not measured separately from genuineness and respect, but these therapist attributes tend to be highly intercorrelated (Elliot et al., 2011; Moyers, Miller, & Hendrickson, 2005). With clients randomly assigned to counselors, Valle compared the relapse rates for therapists who were high, medium, and low in Rogerian skills (see Figure 1). At every follow-up point the risk for relapse was two to four times higher and drinking days were at least four times higher for clients who had been treated by low-functioning therapists. Using Valle’s rating system with a new sample of therapists, Saanio (2002) likewise found that client drop-out from treatment was linked to therapists with lower levels of Rogerian skills.

Figure 1.

Outcomes for Clients of Counselors with Low, Medium, and High Levels of Rogerian Interpersonal Functioning (from Valle, 1981)

Three additional studies have examined the impact of therapist empathy and client substance use by measuring the association between these two variables as they occur in ongoing treatment, without an attempt to control empathy directly or randomly assign clients to different therapists. All of these studies were prospective, with direct measurement of counselors’ skill levels and documentation of client substance use outcomes. One study explored the relationship between therapist characteristics and client drinking directly in a large metropolitan addiction treatment center where clients received standardized cognitive behavioral therapy (Ritter et al., 2002). Clients rated their therapist’s level of empathy, congruence and unconditional regard using the Relationship Inventory, an assessment tool based on the work of Carl Rogers and widely used in psychotherapy research. Clients also rated the therapists’ level of expertness, trustworthiness and attractiveness. Drinking outcomes were assessed 3 months after treatment in multiple dimensions, including the number of negative consequences due to drinking, the degree of physical dependence and the number of alcohol-related psychosocial problems experienced by the client. Both therapist empathy and expertness were associated with improvements in client drinking outcomes in multiple domains.

Pantalon, Chawarski, Falcioni, Pakes, and Schottenfeld (2004) measured therapist empathy during community reinforcement approach treatment sessions for pregnant and postpartum cocaine users. They used a behavioral rating system (Mechanisms of Action Rating Scale: MARS) to capture frequency counts for the occurrence of therapist empathy and responses to resistance during videotape reviews of sessions. Successful client outcome was defined as urine-verified, continuous abstinence from cocaine for three consecutive weeks during treatment. The authors found that therapists with higher rates of empathy had clients with significantly lower rates of cocaine use. Similarly, there was a significant relationship between more positive therapist skills in managing resistance and improved cocaine abstinence for these women. Finally, Fiorentine & Hillhouse (1999) explored the impact of therapist empathy within the context of treatment matching for gender and ethnic characteristics in an outpatient treatment clinic. They measured counselor empathy with a 3 item scale completed by clients that was consistent with a Rogerian definition of empathy (“My counselor understands me,” “My counselor realizes how my experiences feel to me, “and “My counselor understands me even when I don’t express myself”). Outcomes were assessed by categorizing clients as totally abstaining versus any drug use at all in the 6 months after treatment. Clients’ ratings of their counselor’s empathy were significantly correlated with both treatment engagement and abstinence rates. Gender and ethnic congruency, in contrast to counselor empathy, were not associated with higher rates of abstinence in the follow up period.

To place these studies exploring the relationship between therapist empathy and substance abuse outcomes in perspective, we computed effect sizes for each of the studies cited here, where data were available and appropriate for conversion (Table 1). These findings are consistent with a view of empathy as a moderately strong predictor of substance abuse treatment outcomes, with considerable variability across studies. This relationship between empathy and improved client outcomes, though intriguing, cannot be interpreted to mean that empathy causes better client outcomes. Empathy might causing the improvements in client outcome, but it is also possible that a third variable (for example, client motivation) influences both therapist empathy and client outcomes. Randomized, controlled trials in which empathy (or other interpersonal skills of the therapist) are manipulated have not been conducted for both logistical and ethical reasons (Norcross & Wampold, 2011). Nevertheless, the logic of randomized trials is such that the impact of therapist empathy could be evaluated prospectively without having a treatment condition that is explicitly empathic (Miller & Cooney, 1994). Such a secondary analysis could be conducted in any prospective trial of psychotherapies so long as (1) clients are randomly assigned to therapists, (2) there is a sufficiently large number of therapists treating ten or more clients each, and varying in empathic skill level (Project MATCH Research Group, 1998), and (3) reliable session-specific measures of therapist empathy are obtained. Under such conditions, clients would in effect be randomized to different levels of therapist empathy, and a priori hypotheses could be tested regarding a main effect of empathy and its interactions with client characteristics (Longabaugh & Wirtz, 2001).

Table 1.

Relationship Between Clinician Empathy and Substance Use Outcomes

Study	Empathy Rating Method	Outcome Variable	Cohen’s d	Size of Effect
Fiorentine & Hillman, 1999	Client Rated	Drug abstinence	0.31*	Small
Miller, Taylor, & West, 1980	Independent Observers	Alcohol consumption	1.43**	Large
Pantalon, Chawarski, Falcioni, Pakes, & Schottenfeld, 2004	Independent Observers	Cocaine abstinence	1.22*	Large
Ritter, Bowden, Murray, Ross, Greeley, & Pead, 2002	Client Rated	Alcohol consumption	0.18	Small
		Negative consequences	0.33*	Small
		Degree of dependence	0.39*	Small
		Alcohol-related psychosocial problems	0.44**	Medium

Note.

^*p < .05

^**p < .01

All effects in the direction of empathy producing higher results

Empathy and Resistance

Authoritarian confrontation, the obverse of an empathic listening style, has rather consistently been associated in clinical trials with no change or adverse outcomes in addiction treatment (Boardman et al., 2006; Miller & Wilbourne, 2002; White & Miller, 2007). In one study, a single in-session therapist behavior predicted 42% of the variance in clients’ 12-month drinking outcomes: the more the therapist confronted, the more the client drank (Miller, Benefield, & Tonigan, 1993). It appears that, consistent with Rogers’ assertions, providing an accepting and empathic therapeutic style facilitates therapeutic relationship and positive change, whereas a confrontational style does not.

In a classic study, Patterson and Forgatch (1985) used an ABAB design to evaluate the impact of two contrasting counseling styles: one that emphasized providing information (Teach) and advice (Direct), and one that focused on reflective listening. Within client sessions, the same therapists alternated between these styles on signal every 12 minutes. Client resistance during the session was measured via a reliable behavioral rating system (Chamberlain, Patterson, Reid, Kavanagh, & Forgatch, 1984). Client resistance increased and decreased as a step function in response to counseling style. Teach/Direct increased client resistance by 70% in contrast to empathic listening. Resistance dropped back down with resumed listening and jumped back up with a return to Teach/Direct.

An empathic approach may lower resistance to potentially threatening material (Campbell & Babrow, 2004). In a randomized trial comparing therapist styles with problem drinkers receiving feedback regarding the severity of alcohol-related assessment results, client resistance responses were 70% higher with directive as compared to client-centered counseling (Miller et al., 1993).

Can Counselors Be Pre-screened for Empathy?

It appears that therapist empathy can predict meaningful proportions of variance in addiction treatment outcome (e.g., Miller et al., 1993; Valle, 1981) that are an order of magnitude larger than the between-treatment differences typically observed in clinical trials (Imel et al., 2008) and typically fall within the range of what addiction treatment providers regard to be a clinically meaningful effect (Miller & Manuel, 2008). In psychotherapy research more generally, therapist empathy may account for as much or more outcome variance than therapeutic alliance or specific intervention (Bohart, Elliot, Greenberg & Watson, 2002; Imel, Wampold & Miller, 2008). It could be argued that providing accurate empathy in addiction treatment is an evidence-based practice regardless of theoretical orientation and that its absence will reduce the likelihood that clients will change their substance use.

Is it possible to pre-screen therapists for accurate empathy? We know of no paper-and-pencil measure to do this reliably. Even a measure that directly elicits reflective listening statements in writing (Miller, Hedrick, & Orlofsky, 1991) is no guarantee that clinicians actually do this in practice. Clinician self-descriptions of their own listening skills are simply unrelated to actual skillfulness as rated from practice samples (Miller, Yahne, Moyers, Martinez, & Pirritano, 2004). It is possible, however, to observe directly a counselor’s ability in empathic listening during the hiring process. Just such a procedure was utilized in the Combine study, which employed 40 therapists to deliver a state of the art behavioral treatment for alcohol dependence that relied heavily on motivational interviewing, and thus on empathic listening skills. Faced with the need to quickly recruit and rigorously train therapists in a complex treatment, we decided to pre-screen therapist candidates for empathic listening skills. We developed a procedure in which candidates submitted a work sample of themselves facilitating a discussion with another person speaking on one of two topics: “what it was like growing up in my home,” or “how I came to work in this field.” The length of the conversation was at least 20 minutes. The candidate was told to listen empathically without trying to solve problems or give advice. These work samples were then reviewed using a rudimentary coding system to quantify open and closed questions, simple and complex reflections, and confrontations. Counselors with a ratio of at least 1:1 for reflections and questions (that is, at least one reflection for every question) earned a passing score, provided the questions were also at least 50% open questions. Although this bar was low compared to the expertise expected for the actual trial (a 2:1 ratio of reflections to questions is considered an expert level of MI practice) this screening process helped to insure that clinicians recruited to deliver the combined behavioral intervention (Longabaugh, Zweben, LoCastro, & Miller, 2005; Miller, 2004) in this study were equipped with foundational skills to learn quickly and deliver the treatment well. Of 68 candidates submitting work samples for review, 47 (69%) received a passing score at the first submission. After another attempt, an additional 10 candidates were approved, with 11 (16%) never meeting criteria for hiring in the Combine study (Miller, Moyers, Arciniega, Ernst, & Forcehimes, 2005).

Would Empathy Screening at Hiring Be Related to Counselor Performance Later?

Is there any evidence to support the idea that pre-screening for empathy would lead to higher levels of empathy in later counselor functioning? Although there are no prospective hiring studies to address this question, our own training research yields an interesting analog. We conducted two randomized, controlled trials to train substance clinicians in the use of motivational interviewing (Miller et al., 2004; Moyers, Martin, Houck, Christopher & Tonigan, 2009). In both studies clinicians 1) submitted a baseline tape of themselves doing substance disorder treatment in their usual fashion, 2) then received training in motivational interviewing, and 3) then submitted work samples of themselves using MI with actual substance disorder clients at 4, 8 and 12 months afterward. All of these samples were coded using an objective rating system measuring therapeutic empathy as a global characteristic on the same 7 point Likert scale (Miller, Moyers, Ernst & Amhrein, 2003; Moyers, Martin, Manuel, Hendrickson & Miller, 2005). This allowed us to compare the clinicians’ baseline levels of empathy (before training and without their knowing what characteristics or skills we were measuring) with those that were observed later. The results of both of these studies have been published showing substantial training effects with and without various forms of training enrichment such as consultation and feedback. To address the question of whether empathy pre-screening might predict later counselor performance, we performed new analyses of clinician scores on the global empathy ratings scale (n= 207). To simulate an employment “pre-screening” we selected clinicians who scored a 6 or 7 on the empathy measure (“successful applicants”, n= 32) and those who scored a 1 or 2 (“unsuccessful applicants”, n= 30). We then used regression analyses to predict empathy ratings at the 4 month follow up point and found that “pre-employment screening” empathy ratings were a significant predictor of clinician empathy in later therapy sessions (β = 1.49, p < .001, 95% CI (.75, 2.2), R² = .270, F(1,44) = 16.3). It is worth noting that these sessions were gathered from a variety of settings including hospitals, primary care clinics, addiction treatment offices, methadone clinics, domestic violence programs and employment screening programs. All of the tapes were rated by coders who were masked to the time point they were coding, and published reliabilities for the global ratings were generally in the good to excellent range. These data support the proposition that even a relatively gross sorting of clinicians into a pass or fail category regarding their baseline empathy skills will predict later performance in actual treatment sessions.

A Modest Proposal

From the evidence to date, it appears that empathy is a reliable predictor of counselors’ success in treating at least alcohol use disorders. In fact, empathy may exert a larger effect in addiction treatment than has been generally true in psychotherapy, accounting in some studies for a majority of variance in client outcomes (e.g., Miller et al., 1993; Valle, 1981). Najavits and Weiss (1994) observed that in addiction treatment, outcome differences among therapists may be larger than those in psychotherapy more generally. Why would this be so? One possibility is that historically in American addiction treatment, low-empathy confrontational counseling has not only been an acceptable therapeutic style, but has at times been lauded as essential (Janzen, 2001). Tactics such as shaming and demeaning, head-shaving, sarcasm, shouting insults in a client’s face, and “attack therapy” (Yablonsky, 1965), all of which were once regarded as acceptable if not essential in addiction treatment, would be unusual if not outright malpractice in the treatment of virtually any other mental disorder. Use of such practices continues, though no longer representative of addiction treatment, and this legacy has probably contributed to broader variance in provider empathy in this field than would be true in behavioral health more generally. Whenever the range of a variable is restricted its predictive power tends to be diminished. For example, whereas Graduate Record Examination (GRE) scores do predict the ability of applicants to function in post-graduate training (Kunzel & Hezlett, 2007), they are less powerful in predicting grades once students have been admitted to graduate school and the range of GRE scores is compressed. If empathy is an important determinant of treatment outcomes and if addiction treatment providers manifest a wider range of empathic skills relative to psychotherapists, then empathy would be expected to be a more robust predictor of outcomes in addiction treatment than in psychotherapy more generally, precisely because of the presence of more low-empathy counselors.

Outlier therapists with outstandingly poor client outcomes are often found in addiction treatment studies (Luborsky et al., 1985; McLellan et al., 1988; Miller et al., 1980; Project MATCH Research Group, 1998; Valle, 1981). Available evidence links implicates low empathic skill as a marker of this outlier status (Miller et al., 1980; Valle, 1981). From the ethical minimum of “First, do no harm,” it is reasonable to screen for and teach accurate empathy as a key therapeutic skill regardless of theoretical orientation (Norcross & Wampold, 2011). We know of no therapeutic approach where low empathy has been linked to better outcomes in any area of health care. It is both possible and ethically sensible to screen potential providers of addiction treatment services for skillfulness in accurate empathy as an important general factor impacting client outcomes. Of “evidence-based practices” currently being promoted, this seems to us to be one of the most promising to improve outcomes and prevent harm in addiction treatment. In contrast to the notion that empathy represents error variance or that it is unscientific to explore its impact on client improvement, it is our contention that empathy represents a critical component of successful treatment that merits both scientific investigation and greater emphasis in treatment endeavors.