We are looking at 3 common possibilities for modeling star joins in a single collection to mimic OLAP / data mart models for analytics in MongoDB:

1. Fact grain embedded in the array
{
"patient_id" : 123,
"demog" :

{ "birth_dt" : ISODate(“1977-05-12”), "gender" : "F", "state" : "MI" }

,
"PCP" : "Dr. John Smith",
"claim_history" : [

{ "claim_id" : 1456, "claim_dt" : ISODate(“2009-11-19”), "Prim_Dx" : "323.9", "DRG" : 127, "net_pay_amt" : 432.12,"LOS" : 1 }

,

{ "claim_id" : 2456, "claim_dt" : ISODate(“2009-12-01”), "Prim_Dx" : "326.1", "DRG" : 253, "net_pay_amt" : 112.52, "LOS" : 4}

,

{ "claim_id" : 4456, "claim_dt" : ISODate(“2010-01-06”), "Primary_Dx" : "323.9", "DRG" : 322, "net_pay_amt" : 552.75, "LOS" : 1 }

]
}

2. Simple Array of Values (creates a M:M mapping OLAP problem)
{
_id : “joe”,
age : 23,
gpa: 3.23,
classes : [ “BUS252”, “MUS101”, “CS220” ]
}

3. Array contains a collection of references to other documents. How often do you feel that we’ll see references like this?

{
_id : “B7XF873”
"username" : “joeUser”,
"Full Name" : “Joe User”,
"Age" : 25,
“Occupation” : “Baker”
…
“Following” : [ “LLXF321”, “ADE3F44”, ... ]
}

Are these the most common models that you see or is there a recommendation from Mongo in building analytic models in Mongo?