We hypothesized that scientific misconduct (data fabrication or falsification) is goal-directed behavior. This hypothesis predicts that papers retracted for misconduct: are targeted to journals with a high impact factor (IF); are written by authors with additional papers withdrawn for misconduct; diffuse responsibility across many (perhaps innocent) co-authors; and are retracted slower than papers retracted for other infractions. These hypotheses were initially tested and confirmed in a database of 788 papers; here we reevaluate these hypotheses in a larger database of 2,047 English-language papers. Journal IF was higher for papers retracted for misconduct (p
< 0.0001). Roughly 57% of papers retracted for misconduct were written by a first author with other retracted papers; 21% of erroneous papers were written by authors with >1 retraction (p
< 0.0001). Papers flawed by misconduct diffuse responsibility across more authors (p
< 0.0001) and are withdrawn more slowly (p
< 0.0001) than papers retracted for other reasons. Papers retracted for unknown reasons are unlike papers retracted for misconduct: they are generally published in journals with low IF; by authors with no other retractions; have fewer authors listed; and are retracted quickly. Papers retracted for unknown reasons appear not to represent a deliberate effort to deceive.