Bootstrap in models with dummy variables I have applied the bootstrap technique in a multiple regression model where some dummy variables are included, 
is there anything special about the treatment of dummy variables?
 A: When you perform bootstrapping for a multiple linear regression model of the form:
Y = beta0 + beta1*X1 + ... + betap*Xp + epsilon,  (1)

you can bootstrap either the residuals associated with that model or the cases, depending on whether your predictor variables can be treated as arising from a fixed design or a random design.  (This response assumes the model errors are independent and follow a normal distribution with mean 0 and unknown standard deviation sigma.)
An example of predictor which can be treated as arising from a fixed design is temperature - if you conduct an experiment to relate temperature to reaction time for a chemical reaction and you decided to include only the temperatures 10 degrees Celsius, 15 degrees Celsius and 20 degree Celsius in your experiment, Temperature can be treated as arising from a fixed design. If, however, you didn't control the value of temperature but rather measured it alongside with reaction time, than temperature can be treated as arising from a random design.
Bootstrapping residuals (Fixed Design)
If you bootstrap the residuals, each residual will be computed as the difference between the outcome y and the fitted value b0 + b1*X1 + ... + bpXp, where some of the Xi's are dummy variables (i = 1, ...,p), where bi is the estimated value of betai, i = 1,...,p.
Bootstrapping cases (Random Design)
If you bootstrap the cases, you will essentially have to select n rows at random with replacement from the vector of Y values (in column format) and from the design matrix X for your model, where n is the number of observations included in the model. This design matrix will have a column of 1's for the intercept term and then a column of X1 values, a column of X2 values, ..., a column of Xp values. Once this selection is done, you will need to fit the model in equation (1) to the data, which includes some dummy variables. 
                             ---//---

So it doesn't look like you need to pay special attention to the dummy variables, as long as they are included in your model when bootsrapping residuals OR included in your design matrix when bootstrapping cases.
If the dummy variables are used to encode the effect of a categorical variable in your model, then make sure you use k-1 dummy variables to encode the effect of a categorical variable with k categories. The rest should look after itself.  
If you specify the design matrix X "by hand", make sure it includes every single dummy variable included in your model (e.g., it includes the k-1 dummy variables used to encode the effect of a categorical variable with k categories). 
