Second-order latent growth models with shifting indicators (the shifting indicators model) allow longitudinal researchers to add or drop items to develop models that closely represent prevailing developmental theory. To date, however, published research evaluating the performance of the shifting indicators model has been minimal. Simulation methods were used to generate data where all indicators were present at all time points. Data for selected indicators were then deleted to create models with shifting indicators. The performance of shifting indicators models was compared to the original model with all indicators present. The number of shifting indicators per factor, the number of measurement occasions with shifting indicators, the magnitude of the factor loadings of the shifting indicators, and sample size was manipulated. Samples were drawn from multivariate normal populations, and for each cell 1000 replications were obtained. The results of the study indicated that the performance of the shifting models was quite similar to the performance of the models with all items included at each time point. Nonconvergence and inadmissible solutions were rare. Mean values of relative bias in the growth parameter estimates and their standard errors did not exceed .05 and .1, respectively, for all cells with sample size exceeding 250. The shifting indicators models were slightly less efficient than the models with all indicators present, but the difference was small. The investigation into model fit presented one potential caution. Having fewer items per factor was associated with better measures of fit, especially when the sample size was 250. This result indicates that model fit, as measured by chi-square and related fit indices, may be improved simply by dropping items from the model. The overall findings were promising and support the continued study of the shifting indicators model. A demonstration of the shifting indicators model with real data reinforced these findings.