use spss to analyse a cluster randomized data I have data from a cluster randomized trial, that is, each subject in the same cluster has received the same treatment.
Do chi square test and t-test need some adjustments?
More precisely, which analysis have I to carry out in Spss to compare frequency in contingency tables and to compare means for quantitative variables? 
 A: There are a few methods approaches to handling analyses of these types of experiments. I cannot guide you through how to conduct analyses in SPSS since that's beyond the scope of this forum, but you can consult the documentation (it's quite good).
Aside from the relevant question, the most important caveat here is probably the numbers of clusters. Also consider whether there are nested clusters (e.g. repeated measures on a student in a school for which other students are measured). If the number of clusters is small, you can use fixed effects to account for site effects. This is just a factor variable for site, structurally a 0/1 indicator for each site (aside from an arbitrary reference site). This allows predicted mean response levels to differ between sites.
If there are large numbers of sites, a mixed effects model or GEE can be a better approach. Mixed models can use random effects for clusters, (something like family members within a household), to account for a random distribution of a very large number of effects (too many to conceive of using fixed effects, as above). The interpretation of the model estimated effects would then be comparisons between individuals belonging to the same cluster. GEE, alternately, averages out all those random effects and gives you population averaged effects, so your interpreted effects are not restricted to comparisons between individuals in the same cluster.
Often times, when interventions are applied at a community level, e.g. focus groups, outreach, we are curious whether the intervened sites themselves show a demonstrable difference from control sites. A very similar approach is a hierarchical model which is effectively two steps: aggregate results within sites using means and estimated standard errors of those means. Then perform a test of differences between treated and control sites using inverse variance weighting. This accounts for the numbers of individuals between sites (and the variance of their measurements) but allows you to compare differences on average between sites receiving treatment versus those receiving control.
