Dear Edita,

Yes, as an approximation to random sampling; in many cases there is not other solution. A possibility is to use stratified sampling in two strata (12-17 years old, and 18-21 years old, as the two strata; or with a socioeconomic boundary of strata, for example, with income inferior to, and superior to a fixed income). Also any combination of both procedures of stratification too, for example with four strata ("2 strata for age" x "2 strata for income").

Sometimes it makes sense to use purposive sampling when you can make very specific predictions from a hypothesis, and sometimes you just don't have much choice than to use a purposive or even convenience sample. However, if you are trying to make inferences about a whole population from your sample, you'd get much more meaningful results by using some form of probability sampling. In this case, a stratified random sample would likely be a good choice if you can plan it in advance to ensure that there will be adequate sample sizes in each stratum. How are you recruiting the research subjects? Can you select samples from a database of some kind? Or do you have to recruit them on the street or through a school system? The recruitment aspect will make a big difference as to how you could do this and even to how you define your population.

Very few people use true "convenience" sampling because that implies taking just about anyone. At a minimum, you already have eligibility criteria, and you have set up the possibility of systematic comparisons, so you are doing purposive sampling by definition. What you don't have is random selection within your subgroups of interest, which means your statistical results will not generalize to the population.

okay, thank you so much for the answers! I really appreciate it.

Edita -

You mentioned "correlational research."  Does that mean you are looking at a relationship between groups which may be expressed as a linear regression?  If so, then you could make estimates and variance estimates with various subgroups within your strata to see if there appear to be unexplained differences.

Whatever kind of sampling you do, it will be improved by good stratification.  To do a good stratification (random sampling or not), you need auxiliary information to categorize data. 

If you also have continuous (rather than discrete) auxiliary data on each population member of a stratum, you can do regression for another purpose: you can use those continuous auxiliary data as regressor data to estimate the portion of the population total included in the sample for each variable of interest.  As examples of using linear regression with a zero intercept to estimate totals for variables of interest in Official Statistics with missing data, there is the following: https://www.researchgate.net/publication/319914742_Quasi-Cutoff_Sampling_and_the_Classical_Ratio_Estimator_-_Application_to_Establishment_Surveys_for_Official_Statistics_at_the_US_Energy_Information_Administration_-_Historical_Development

Note, FYI: Random sampling without auxiliary data generally produces far less accurate results.

I do not know your exact goals, but perhaps something of the above might be of some use to you. 

Cheers - Jim

PS -  If you possibly can randomize selection within the groups you described in your study, it sounds like that could be quite helpful here, so best think about that first.  -  Best wishes.

sounds more like purposive sampling would be more helpful since you have a criterion "participants will be adolescences which are from middle low SES family" , which binds your choice to a limited category.