[lbo-talk] green consumers: thieving pricks

[Miles Jackson]

Doug Henwood wrote:
>
> On Mar 19, 2010, at 11:23 AM, Miles Jackson wrote:
>
>> I'm making no heroic assumptions; I'm just passing along what every 
>> statistics student learns in their first course on probability.  If 
>> you randomly assign people to two groups, the potential confounds are 
>> also randomly distributed across the groups, and it is not plausible 
>> to argue that the groups systematically differ.  Really, it's 
>> Probability 101.  Random assignment to experimental condition is a 
>> very powerful and rigorous technique to control confounds here 
>> (including any potential confounds you and I are not clever enough to 
>> imagine!).
>
> You could argue that the group sizes are too small for true randomness 
> to prevail. What's the thinking on this in psychology? How big a group 
> do you need?
>
> Doug 
There is no strict cutoff.  About 25 to 30 per experimental condition is 
typically enough to randomly distribute potential confounds.  Dinky 
group sizes (say, 10 per condition) could be a problem.  It's like 
flipping a coin; it's easier to get an unequal distribution of heads and 
tails with a smaller number of flips.

Miles