This menu allows you to fit the variance shift outlier model (VSOM) to detect outliers in a REML analysis.
Residual term to check for outliers
The dropdown list contains the residual strata in the REML model. Select the strata that is to be checked for outliers.
Display
This controls what output is printed from the analysis:
| Outliers | A list of possible outliers |
| False discovery rate | The false discovery rates for the potential outliers, which are estimated from the distribution of p-values calculated from the t-statistics using an asymptotic model, See the FDRMIXTURE procedure for more information on the estimation of false discovery rates |
Plot
This controls what plots are displayed from the analysis:
| Index plots | The estimated statistics from the analysis plotted against their index (order) in the data set |
| Residual plot | The usual residual plot with outlying points coloured by their outlier status |
Components in the index plot
The index plot can have three components plotted:
| Omega | Variance shift as a ratio to the residual variance |
| Sigma squared | Estimated residual variance under VSOM |
| Statistic | The statistic selected in the Methods section |
The graph can be given a Title by entering the text in that field. If no title is given, then default titles will be used. Use a single space in the field to suppress titles.
Methods
This controls what plots are displayed from the analysis:
Calculating statistics
The deviance associated with a variance shift for a particular unit in the strata can be calculated using 3 statistics:
| t | Uses the squared t-statistics (i.e. squared standardized residuals) to approximate the change in likelihood (default); this is the fastest approach |
| Partial likelihood | The partial likelihood which approximates the change in likelihood, where the baseline model parameters are held fixed, and only the extra variance component for each unit is estimated; this is much faster than re-estimating the full model |
| Full likelihood | The full likelihood calculated by refitting the full model with the added variance term for each unit; this can be very time-consuming |
Constrain variance components to be positive
This constrains the variance components in the model, including the variance shift for the units, to be positive.
Calculating thresholds
The thresholds used to decide if an outlier is significant at the 5%, 1% or 0.1% level can be calculated two ways:
| Bootstrap | Uses parametric bootstrap samples, with the variance components in the baseline model, to calculate the thresholds from the percentiles of the order statistics |
| Approximate | Uses the asymptotic distribution to calculate the thresholds |
Number of samples
This option specifies the number of bootstrap samples that are performed, default 499 (as well as the “null” model where the data keep their original values).
Seed
Specifies the seed for the random number generator used to make the bootstrap samples; default 0 continues from the previous generation or (if none) initializes the seed automatically.
Save
Use these options to save results from the VSOM analysis. After selecting the Results box, you need to type the identifier of a pointer into the In: field. The elements in the results pointer are given in the VSOM help.
Display in spreadsheet
The saved results will be displayed in a new spreadsheet.
See also
- VSOM procedure to calculate the variance shift outlier model.
- VPLOT procedure for plotting residuals.
- FDRMIXTURE procedure.
- Linear Mixed Model menu.
- Initial Values for specifying initial gamma.
- Further Output for obtaining additional output after fitting a model.
- Save for saving the results from a REML analysis.
- Save REML results in a spreadsheet.
- Residual Plots for generating plots of residuals.
- Means Plots for generating plots of one- or two-way tables of means.
- REML directive for command mode use of REML, with additional options to
control the algorithm and for more sophisticated analyses. - VCOMPONENTS directive for further information about fixed, random, and spline model terms.
- Linear Mixed Models (REML) – Correlated Errors for setting up covariance models.
- REML Power calculations dialog for calculating the power of a REML analysis.
- REML Permutation Test dialog for calculating the significance of a random term using a permutation test.