平衡量表是指同时包含正向表述题目和反向表述题目的量表，研究者通常采用单维模型和对称的双因子模型来拟合平衡量表。但尚未有研究在平衡量表情境下系统地比较单维模型和双因子模型的表现，且对称的双因子模型在拟合时存在较高异常率。本研究拟通过系统的模拟研究比较各测量模型在平衡量表上的适用性，并尝试引入双因子模型的拓展形式双因子-（S-1）模型来解决拟合问题，同时创新性地纳入结构模型来辅助测量模型选择。

研究一通过模拟研究，在不同因子负荷和结构系数条件下，比较三种模型的拟合效果和预测效度。使用模型拟合的BIC、CFI、TLI、SRMR、RMSEA和异常率作为模型拟合效果指标，使用因子分数的标准化估计值和真实值的误差均方根作为比较模型拟合准确性的指标，使用结构系数和结果变量残差的标准化估计值和真实值的相对误差均方根和相对偏差作为比较模型预测效度的指标。结果发现几乎所有模拟条件中单维模型的拟合效果均未符合标准，双因子-（S-1）模型和对称的双因子模型的拟合效果均符合标准，但对称的双因子模型的异常率远高于另外两种模型；此外，当双因子-（S-1）模型中的局部因子是独特效应更强或共同效应更弱的观测变量所负荷的局部因子、或是结构系数更大的局部因子时，其模型拟合准确性和预测效度均优于对称的双因子模型。模拟研究证明双因子-（S-1）模型在平衡量表模拟情境下的优越性，并对双因子-（S-1）模型中局部因子的选择提出相应的建议。

研究二以480名高校学生为研究对象，采用Rosenberg自尊量表和简版韧性量表作为目标平衡量表，自杀行为作为结果变量，分别拟合三种模型。结果发现，对于两个平衡量表，将反向表述题目负荷在局部因子上的双因子-（S-1）模型均表现出了比单维模型和对称的双因子模型更好的拟合效果；自尊特质、心理韧性特质和两个反向表述因子均与自杀行为有显著负相关。研究二进一步验证了在真实情境下，采用双因子-（S-1）模型拟合平衡量表的有效性。本研究的发现为平衡量表的应用提供了有效的模型选择依据。

The mixed-format scale refers to a scale that contains both positive and negative formulated items. Researchers usually use unidimensional model and symmetric bifactor models to fit mixed-format scales. However, there has been no systematic comparison of the performance of unidimensional model and bifactor models in the context of mixed-format scales, and symmetric bifactor models have a high abnormality rate when fitted. This study intends to compare the applicability of various measurement models on mixed-format scales through a systematic simulation study and attempts to introduce an extended form of the bifactor-(S-1) model to address fitting issues, while innovatively incorporating structural models to assist in model selection.

Study 1 compared the fitting effect and predictive validity of three models under different factor loadings and structural coefficient conditions through a simulation study. The BIC, CFI, TLI, SRMR, RMSEA, and the abnormality rate of model fitting were used as indicators of the model fitting effects The root mean square error (RMSE) between the standardized estimated values and the true values of factor scores was used as an indicator to compare the accuracy of model fitting. While the relative root mean square error and relative bias of standardized estimation values and true values of structural coefficients and residuals of outcome variables were used as indicators for comparing model predictive validity. The results showed that the fitting effect of the unidimensional model did not meet the standard in almost all simulation conditions, while both the bifactor-(S-1) model and the symmetric bifactor model met the standard, but the abnormality rate of the symmetric bifactor model was much higher than the other two models. In addition, when the specific factor in the bifactor-(S-1) model is loaded by observed variables with stronger unique effects or weaker common effects, or when the specific factor has larger structural coefficients, the model exhibits superior fitting accuracy and predictive validity compared to the symmetric bifactor model. The simulation study demonstrated the superiority of the bifactor-(S-1) model in the context of simulating mixed-format scales, and provided corresponding suggestions for selecting specific factors within the bifactor-(S-1) model.

Study 2 used 480 college students as participants, fitting three models using the Rosenberg Self-esteem Scale and Brief Resilience Scale as the target mixed-format scales and suicidal behavior as the outcome variable. The results showed that the bifactor-(S-1) model, which loaded negative formulated items on a separate factor, demonstrated better fitting effects than the unidimensional model and symmetric bifactor model. Self-esteem trait, resilience trait and negative statement factor were all significantly negatively correlated with suicidal behavior. Study 2 further validated the effectiveness of using the bifactor-(S-1) model to fit mixed-format scales in real situations. The findings of this study provide effective model selection criteria for the application of mixed-format scales.