Can a convenience sample ever be considered truly cross-sectional?

#1
I'm having a debate over terminology with a colleague. We've done a small census survey (n<50) of residents and staff at a local out-patient clinic. Variables are knowledge/attitudes about resident education practices. We are making no assertions about generalizability to other clinics. Is it appropriate to call this a cross-sectional design? I would argue, no, that the term cross-sectional implies (or requires) that a sampling procedure has been used to identify and select participants so that inferences can be made from your sample to the population.

I would just call it a census survey; she wants to call it a cross-sectional design. To me, there is nothing cross-sectional about it. Various definitions in both books and web suggest that she may be technically correct, but this seems to me a misuse of the term.

Opinions, or clear evidence one way or the other?
Many thanks,
Bret
 

noetsi

Fortran must die
#2
Different people use terms differently. To me cross sectional has nothing to do with external validity or sampling at all. It means that you are not measuring results over time - that is X is assumed to influence Y only at one point in time (unlike time series the other form of analysis where X has immediate and long term impact over time). I have never seen cross sectional used to apply to sampling or external validity before.
 
#3
Thank you for your reply.

"Different people use terms differently." Yes, which is what I am trying to avoid and thus seeking guidance from this forum.

You say "I have never seen cross sectional used to apply to sampling or external validity before." Just about every definition of "cross-sectional" that I have looked at contains some reference to a population. The term itself suggests a sampling process: that's the whole idea of a "cross-section" isn't it? It is a temporal quality, but it is also a sampling issue.

If, for example, I randomly selected and collected data from one classroom in a school, there is no way I could claim that data was cross-sectional. It isn't a "cross-section" of anything. Any results are relevant only for that one classroom. For the term to have any meaning, it seems to me it must go beyond simply data collected "at one time."
 
#4
Like noetsi, I too have seen different cross-sectional researches with different sample types such as convenient, sequential, random-sequential, simple random sampling, other types of random sampling, clustered sampling, etc. A cross-sectional study reflects somehow on the population type (i.e., not caring about the time element as noetsi pointed out), but does not care about the details of the sampling. A random sample is needed for proper generalizabiliy, but it has nothing to do with the study type being or not being cross sectional.

If, for example, I randomly selected and collected data from one classroom in a school, there is no way I could claim that data was cross-sectional. It isn't a "cross-section" of anything. Any results are relevant only for that one classroom.
You are mixing two different topics. The sampling is only random or non-random (here random). But a random sample can be used in a cross-sectional study or other types of studies. Inferring one (study design) from another (sampling method) is not even the case in the quoted example. :)

It isn't a "cross-section" of anything. Any results are relevant only for that one classroom.
Cross-sectional is a terminology for a specific type of study. It is named after some sort of sampling (I think against longitudinal design) but it is not depending on the exact meaning of the term. For example you still call a parked car an automobile right? Then we should call a car different names when it is moving or parked? Or should we call an aircraft an earthcraft when it is on the runway or a trashmetal when it is parked (because when it is parked, it is of no use but trash!)?!

Besides, this is not true that any results are only relevant to that single classroom. For example, you count the number of wings of 10 chickens in a chicken house. Your results indicate that chickens have 2 +- 0 wings. Can't you say that "I think all chickens in the world are likely to have 2 wings each?" Or do you say that "my result is relevant only to this chicken house and it does not tell anything else at all about other chickens in the world"... True answer: not only that sample tells you some relevant information about all the chickens in the world, but also your result is relevant to all birds in the world. The same applies to your "classroom" example.

So yes your class sample *is* a cross section of humans at a specific age in a certain type of school and having other properties.
 

noetsi

Fortran must die
#5
Almost all statistical analysis is assumed to estimate a population through a sample. Rarely if ever are you just concerned with the sample itself - you want to generalize to the larger population. So using that distinction to define cross sectional does not make a lot of sense. Whether it is cross sectional or time series it is assumed to apply to a larger population which you are estimating through a sample. The difference is with cross sectional data, unlike time series, you are only estimating immediate effect (or perhaps you could argue that the effect of X on Y does not change over time).

Part of the problem here is that statistics historically only addresses internal validity. Generalizability, external validity which is the realm of sampling, is rarely considered in discussions of statistics.