Many researches caught sight of the relationship between warmth and competence of stereotype content model. Their negative relationship comes from a comparative context of two social targets with the hypothesis that stereotype is stable. Few theories give exact explanations for this, while a dynamic compensation of one group target may be exist, so this article aims to explore that in-group members may change stereotype content judgements from the state of favoritism into compensation. Undergraduates of different majors were recruited from a university in Tianjin. Control group had 128 participants (53 males, 68 females, 17 who did not indicate gender; mean age = 20.1, SD = 2.17), they completed paper-and-pencil version of the questionnaire of stereotype content about undergraduates. Another 428 participants of experiment group (176 males, 237 females, 15 who did not indicate gender; mean age = 20.0, SD = 2.02) had to follow the procedure of serial reproduction method firstly. Experiment group participants were divided into three groups by different stereotype-inconsistent information assigned randomly. Three short essays were made up of and contained negative judgements about either undergraduates’ competence, sociability or morality respectively. Serial reproduction method is that the first participant of each chain read the original stimulus twice at their own pace. After a distractive task for 5 minutes-drawing a floor plan of their own house, Participants were asked to recall the essay as much as possible without a deadline. Their reproductions were typed verbatim and printed for subsequent participants, who repeated all the tasks later. Each reproduction chain contains four participants. There were 37 chains with negative competence information, 36 chains with negative sociability information and 34 chains with negative morality information. There was none significant difference among the means of control group’s three dimensions scores (competence M = 3.57, sociability M = 3.50, morality M = 3.45, F (2, 254) = 1.45, p = .236). This fitted the hypothesis of in-group favoritism. But the experiment groups showed two kinds of compensation through independent-samples t test. As predicted, participants strongly endorsed undergraduates with much less competence (M = 3.36), and more sociability (M = 3.70), but morality kept the same level after reading competence inconsistent information. It showed a compensation between competence and sociability. When participants were simulated by negative morality information, the compensation was less morality (M = 3.19) and more sociability (M = 3.71), but equal competence (M = 3.63). Another kind of compensation came from the sequence of negative sociability information that only the score of competence became higher. At most positions of the reproduction chain, the above-mentioned compensation almost made the difference among the three dimensions. But they disappeared along the reproduction chain gradually with none significant difference among the positions or the quantities of reproductive issues. These results partly supported the hypotheses that compensation could come from in-group by a dynamic way and the stereotype is not always stable. This result retained the meaning of comparative effect along the time line, not for two targets, but only one group at two different stations with stimulus or not. Meanwhile, the system justification theory could also support the compensation by offering a reasonable knowledge about the unbalance in stereotype content. But the mechanism of the in-group compensation is still unclear, perhaps it contains two steps, unbalance state after comparing first and making up then. So it will be very useful to figure out the progress of compensation. Finally, the difference between sociability and morality is so obvious that it’s not reasonable to combine them as warmth.