 |
 |
Dear SPMers, I have some troubles in understanding this quotation from chapter 12 of "Human Brain Function 2nd edition": "A balanced design is one in which all subjects have identical design matrices, and is a requirement for the SS [Summary Statistics] approach to be valid. If there are, say, two populations of interest and one is interested in making inferences about differences between populations then a two-sample t-test is used at the second level. It is not necessary that the numbers of subjects in each population be the same, but it is necessary to have the same design matrices for subjects in the same population ie. balanced designs at the first-level." I would like to get your point of view on how to deal with subject-level designs in these studies : 1) We have 4 sessions per subject (with 1 task and 1 condition each). The HRF is modeled by the Informed Basis Set thus there are 3 regressors per task plus 1 regressor per session. Most of the subjects performed the four tasks (the design matrix looks like matrix4tasks.png attached), however for some subjects, 1 or more tasks are missing (either because the subject did not perform the task or because motion was too severe) and therefore the design matrix is different (cf. for instance matrix2tasks.png). At the second-level we look at each condition in a separate model and also at the combined effect of all tasks (that is why we initially chose to model all tasks in the same design at the first level). For the combined-task analysis the design matrices are identical at the subject level since we keep only the subjects who performed all tasks. However, when we deal with the effect of one particular task, the design matrices at the subject-level are not all the same. I am wondering whether this is an unbalanced design and if I should instead consider a separate design matrix for each task at the first-level ? 2) The second example I can think of occurs when the conditions depend on the subjects' behaviour. For example, if the subjects are asked several times to make a choice between two inputs we could create two regressors to look at the response when the subjects select a given input. Then, again the matrices won't be exaclty the same (since the subjects won't make the exact same choices). I wonder if this a case of unbalanced design or if it is valid to perform a second level analysis to make inference on the population ? Thank you for your help, Best regards, Camille -- Camille MAUMET Unité/Projet VisAGeS U746 IRISA-INRIA Rennes Campus de Beaulieu 35042 Rennes Cedex France [log in to unmask]