The SIMILE Server allows you to retrieve stored models, designs, and public data sets. The models and designs are useful as templates. The data sets are instructional so that you can see how SIMILE works.

The SIMILE Server allows you to generate, store, and retrieve your own models, designs, and data sets.

Once you have retrieved or generated a model/data set , you can run SIMILE to fit the model and estimate the parameters. The fit statistics and estimated parameters may also be stored and retrieved for later use.

PROBLEM DESCRIPTION

Enter a short name for the model/dat set and description of the model and data.

DESIGN

SIMILE first requires a design based on the information factors used and the combinations presented. You may input up to ten factors. Each factor is specified by entering a single letter (e.g., A, B, C) in the first row of the Factor Table, a short label of the factor in the second row, the maximum number of levels of the factor in the third row, and a blocking option the last row. This option will be explained at a later point.

The design is specified by listing a cell number (information combination) and the list of factor variates or the levels of each factor presented. If a factor is not presented (e.g., the information is ommitted or missing), a zero should be entered. The design is entered in the Design Table, one cell per row, with one space between levels.

Since many designs are generated by simple factorial designs and combinaotiral subdesigns, SIMILE allows you to enter the factorial designs as strings of factor letters, (e.g., AB, ABC) in the Subdesign Table. Subjects judge repeated blocks of these combinations, you can enter the number of blocks and SIMILE will generate these blocks in the Design Table.

DATA

The data for each cell in the design is entered as a row corresponding to information combination in each row of the Design Table. If there are multiple subjects, the data for each subject is separated by a space on the same line. Consequently in the data table, rows are within-subjects and columns are between subjects.

The Summary Statistics Table calculates the mean and standard deviation for each cell in the design and enters the overall standard devation across cells.

MODEL SPECIFICATIONS

Information integration models vary in a number of important ways. The Model Specification panel allows for either linear or multiplicative models, adding or averaging, equal weighting or differential weighting, and with or without a weight of an intial impression. In addition, the Model Specification allows the user to specify which subjective values and/or weights may or may not be equal among the factors. For example, Factors A and B might have to do with product quality from two different sources. Consequently, the subjective values for levels (e.g., Low, Moderate, and High) should be the same, but the weights for the two sources might be different.

The exact model specification will result result in a number of parameters. Some of these parameters will be determined by constraints (e.g., the sum of the weights must equal one or the mean of the scale values on one factor must equal the mean on anothers). Often these constraints are very complex. If the number of parameters to be estimated is greater than the number that of allowable parameters given the design, SIMILE will indicate that the determinant of the matrix for the solution of the equations is zero. Once the model is set, SIMILE should determine the number of estimable parameters in the Parameter Input Table.

The subroutine for Model Specification is currently under development. At present, models are hardcoded for each design and model specification.

Finally, there is a option for unit, homogeneous, or heterogenous variances. Unit variance is used when there is only one observation per cell. Homogeneous variance assumed equal response variance across all cells. It provides the most stable estimates. Heterogeneous variance allows for a different variance estimate for each cell. It is perhaps the correct way to proceed, but is the most unreliable. Other variance options between these are actually more interesting. They would assign different variances to the valuation process, the weights, and even the integration process in addition to a homogeneous variance associated with the response function.

PARAMETER SPECIFICATION

Once the number of allowable parameters has been specified, you may input initial, best guess values as well as constraints on their lower limits and upper limits. If iterated values exceed these limits, they are re-initialized. If values continue to go out of bounds, the maximization with terminate. Finally, you must input accuracy levels. These are delta values such that when iterations do not increment parameter values greater than the accuracy values, the program is said to have converged on a solution.

MAXIMIZATION

The default number of iterations is 50, but you may change this value. Once you start the iterations, the results each iternation with be shown in the Fit Window. When SIMILE converges on the solution, the parameter values will be shown in the Parameter Table and the Likelihood Ration values will be shown in their table. The first row gives -2log(L) for the integration model. The second row gives this value for a model that allows every cell to vary freely. The third row gives the likelihood ratio for the integration model relative to the general model. This value is a Chi-Square statistic with degrees of freedom shown in the last column. If it is significant,it means that the model estimates of cell means depart significantly from the observed cell means. If it is not significant, it provide a goodness of fit support for the model.