These 5 take-a-ways are summarised from this LSE blog article - which you really must go and read for the full story - and its summary of work (in the USA) by Jeffrey A. Smith, Miller McPherson and Lynn Smith-Lovin. This work used used 1985 and 2004 General Social Survey data on these demographic characteristics: racial, religion, gender, age and education.
In summary the finding is that the composition of one’s social network is highly contingent on the composition of the population and thus only changes slowly over time as does the composition of the population.
Which means if you want to escape some of the limits of your own social horizon and understand the views of others you'll need to take specific steps to get to know people beyond those like you - particularly as concerns racial background, religious belief, those much younger or older and those with much higher or lower education.
So the 5 take-a-ways from research on homophily (the tendency for people who are similar to know each other) are:
1) We now know more people from different racial backgrounds - but this change is in line with the changing composition of the population and only some 10% of confidant ties are cross-racial;
2) We now know more people from different religious backgrounds - but this change is in line with the changing composition of the population and only some 29% of confidant ties are cross-religious;
3) On gender some 43% (was 40% in 1985) of confident ties are cross gender - as the the population's gender composition has not changed significantly this change from 1985 is thought to reflect men and women occupying increasingly similar roles in society;
4) On age those in their 20s and early 30s are increasingly more likely to interact with each other, but less likely to interact with those who are older;
5) On education - even though income & wealth inequality has increased - those with lower levels of education are socially closer and less likely to have confidants with those with higher levels of education.