I intend to draw a random sample of public bus transport users in some cities in Ghana. However, I am constrained by the fact that a sampling frame of this population is not feasible. How can I still draw this sample so as to satisfy the use of parametric statistical techniques and more importantly be able to generalise the study outcome? Fortunately, a sampling frame exists for bus terminals in the cities. Can I randomly sample the terminals and then use non-probabilistic sampling techniques to sample the respondents, passengers? Will that be a good practice? Help greatly sought.
I would alsway sample some randomly selected passengers (the sample size of the passengers may roughly reflect the utilization of the corresponding line/connection, if this knowledge is available).
The generaliziation depends on what factors you consider to vary in your study. I see considerable possible effects of the season (summer/winter), weather, weekday and daytime (morning, noon, afternoon, night). If you sample at noon, you may miss many travelers going to work in the morning; if you sample during good weather you will miss people using the bus only when the whether is bad and so on
A good practice is to think about all such factors that can have any impact on the people using the bus (and being directly or indirectly relevant to your survey). Then you must make up your mind if you want to dissect these possible impacts or if you just like an "average" answer (then: average with respect to what? - and this what has to be selected randomly as far as it is possible; what is not possible will be a convenience sample, what is still not a big problem when the convenience is not strongly associated with one of the possibly relevant circumstances and if it is clearly communicated).
Hi, Jochen Wilhelm has suggested very comprehensive solution. Let me repeat him that you should try to include maximum variation in your sample in terms of age, education, bus terminal, timing and day of the week and weather etc. to increase external validity of your study. Further, you can also split the sample by any of the above mentioned variables to see if the different groups vary on outcome/dependent variable(s) in your case. However you need to have sizable sample size to cover the apparent heterogeneity in your population.
Thank you, Jochen and Arif, for the responses. They are very indeed insightful and adds further dimensions to this issue as well. However, the question still remains, is probability sampling possible without a sampling frame, especially in the case of PT users? Is there a technique to generate a possible probability sample from this, especially if the aim is to generalise the findings? In this case, a list of clearly demarcated bus terminals exist.
- Badges
- Science topic