a meta-analysis of rehabilitation outcomes research, results recorded by observers who knew patients' treatment status tended to be higher than outcomes recorded by blind observers. 22 Blinding of raters to treatment group is taken for granted in research studies as a guarantee of the integrity of data; grant proposals to evaluate treatment outcomes in rehabilitation are expected to include such safeguards. The precaution of blinding FSM raters has been recommended as a check on potential gaming, l but has rarely been implemented in clinical settings. Self-serving bias, the tendency to attribute one's own successes to personal characteristics, but one's failures to external circumstances, has been noted in research by Macchiochi and EatonJ 7 In their study, therapists who were surveyed regarding their explanations for neurorehabilitation outcomes did not spontaneously mention the single most predictive variable (injury severity), and tended to provide therapist-related explanations for good outcomes while providing non-therapist-related explanations (eg, patient motivation) for poorer outcomes. Such attribution biases could be reflected in FSM scoring patterns in real-world decision making. For example, an up-coding at discharge might result from a self-serving bias regarding the efficacy of therapy provided when the raters are therapists who had participated in treatment. Similarly, a down-coding at admission would make any improvement by discharge seem even greater, as well as providing a self-handicap 23 to explain any failure of the patient to show gains after treatment. Other biases have been described that can lead to distortion and inaccuracy (eg, confirmation bias where initial opinion is subsequently always confirmed and contrary data are ignored). 2Z'24 Staff group discussion of patient data may increase bias, causing opinions to become more extreme either toward improvement or worsening, z5 In treatment or teaching situations a staff member may act according to expectations of the patient/ student, thereby actively contributing to the anticipated resultwhether positive or negative (self-fulfilling prophecy). 26'27 Decisions may also be influenced by the information available in memory to the decision-maker or rater, creating an availability bias) 1 Information is more likely to be remembered and available when it is vivid (because of recency, interest, or emotional charge) than when it is less vivid but perhaps more characteristic of the patient's functioning. Macchiochi and Eaton I7 demonstrated the importance of availability in clinical judgment by cuing raters to potentially important data or omitting cues. When raters were so cued, they did notice and utilize key information otherwise neglected. Information stored through professional training appears to produce some bias; research indicates that disciplinary differences emerge in ratings on FSMs. 28'29 Another potential source of bias is missing data, ie, ratings not marked on FSM items. Study results from a variety of disciplines indicate that severe bias may appear when such gaps in data sets are random; if item ratings are missing in some systematic way, the distortion may be even more serious. 3° These briefly mentioned biases, while worth noting, are bidirectional, ie, thinking may be biased either for improvement or against it. The situation in rehabilitation outcomes seems more likely to stimulate directional bias in favor of improvement. Some of the biases and heuristics mentioned above have been shown to occur undonsciously and in the absence of external