Anytime you are conducting any experiment that makes use of human subjects, it is important to ensure that it is being done with consideration for the subject's protection and well-being. Even though a social experiment is not exposing your subjects to experimental chemicals or substances that could do direct biological harm, it is important to consider what effect the experiment will have on the subject himself. Coercing a subject into doing something he regrets, even if it is only a simulation, can wrack the subject with guilt and doubt about himself. Consequently, it is important to have any experimental design you ultimately conceive approved by your institution's internal review board. Additionally, be sure to always debrief your subject after the experiment that the session was designed to force them to do something they would not normally do, and is not a reflection on their true character.
Most people are taught from a young age by teachers and parents that stealing is wrong. Most people internalize this lesson, either for strictly moral reasons, or for fear of the consequences of being caught. One way to test the effect of social pressure on inducing unethical behavior, such as stealing, is to invite subjects to participate in an experiment. When they arrive, put them in a "waiting room," tell them that they just need to wait until the current experimental session wraps up. Only, they are in the experiment from the time they sit down. Set up confederates posing as other volunteers waiting for the same experiment. Have your "assistant" visibly store a envelope with cash in an unlocked drawer in a desk, then walk out of the room. This will be the cue for your confederates to try and convince the real participant to take some of the money. While your experiment may attract a few natural-born thieves, with an appropriate "n" number of participants, the occasional outlier will not affect your results.
A similar set-up will serve equally well for an experiment on how easily social pressure can induce someone to engage in bullying behavior. After sitting the true participant down in a room with confederates, and having the "assistant" leave the room, have some confederates who have been pre-selected to be "bullies" begin bullying a pre-selected confederate. Have them try to induce the participant to join in, insinuating that he will be socially accepted by the bullies if he does. Each experimental session should stick as closely to the same script as possible, to avoid an additional variable of different participants being swayed by different wording.
Use the same fake waiting room scenario, only this time, a lab coat-wearing confederate will walk into the room carrying a highly technical-looking device. Have him ask the assistant if he can keep an eye on the device -- it could deliver an uncomfortable shock to anyone who touches it. After the lab-coat adorned confederate leaves the room, simulate some kind of urgent need for the assistant to rush into the room where the "experiment" is being conducted. Once the assistant leaves the room, have the confederates who are posing as research participants begin trying to goad the real participant into touching the device that could shock him. The device would be a completely inert prop with no shocking capabilities.