Yes, you can .Following link can be helpful to you 

https://www.socialresearchmethods.net/kb/expsimp.php

https://www.socialresearchmethods.net/kb/statsimp.php

Nisha Chandel- Thank you so much! The materials you provided really help. I was sure that the only post-test design works. However, some of my colleagues challenged me and said pretest is a must. I am looking for further information from Donald Campbell's book. 

Dear Gordon. I wonder how you measure teacher candidates' teaching skills? This will be useful for justifying why you need or do not need pretests.

Mehmet: Your question makes sense. But my focus is on the design rather than measurement. Thanks!

"Posttest-only designs are most valuable when pretesting is not possible or appropriate but the research purpose is to seek causal relationships."

Page106: DePoy, E., & Gitlin, L. N. (2010). Introduction to research: Understanding and applying multiple strategies (4th edition). Elsevier Health Sciences.

----

You can use this quotation. In your case, you can argue that pretesting is not appropriate. If I were you, I would mention that pre-service teachers in both cohorts have enrolled in similar courses or trainings related to teaching skills before taking the course. I believe that this sentence and the reference above might be enough.

Mehmet: Thank you sir!

Your suggestions are pretty helpful. In effect, the reason for not using the pretest is very practical: I forgot to do that. It is impossible to make up now. Certainly, what you mentioned in the above post is what I will write in future. You can see that I mentioned a little bit in my original post. They are very similar in terms of teaching skills and the classes they had taken and the training they were receiving. 

I don't have a citation for you so much as I'm away from the office, but as a reader with a strong statistical background, I'd be leery of causal claims here. You have a good correlation and possibly regression framework, but even the pre-test data would not be sufficient for making causal claims.

I become concerned with regards to the cohort and time differences that are not accounted for in the quasi-experimental design. The professor had an additional year of experience overall and in teaching the course for the second cohort, which could definitely be related to changes. Are there differences between the people in each cohort? Without pre-test data you have no idea if cohort 2016 or 2017 just had better teaching skills to begin with. There are also concerns related to the time of day the class was offered, the weather, distractions during the class, and more.  A proper quasi-experimental design where causal claims may be made would involve using students in the same exact instance of the class exposed to each learning condition and using pre and post test data for this. Another option would be to compare each student to him or herself using a repeated measures design.  If you have any of Andy Field's statistics texts he provides a great section on quasi-experimental designing that would address all of this.

That said, you should do the study. Consider other methods of analysis as well. Perhaps you can run a correlation on this data and make associated claims.  Are there variables that might act as mediators or moderators in this relationship? (i.e. gender, age, baseline teaching in a different area?)  Although I'd avoid causal claims, you have enough experimental in your design to remove any concern about the teaching skills increasing the exposure to collaborative learning.  This would allow use of regression in asking whether collaborative learning helps predict teaching skills.  With use of effect size via Rsquared you could discuss explained variance as an approximation of a causal claim. 

Hope this helps without being discouraging. It's a good study and interesting at that, but causal claims in the social sciences are very difficult. It is often best to add data to a theory that will help support a future causal claim unless you have an extremely strong quasi or just experimental design. If this is an already well-supported claim, you may use your results to support a potential causal claim in your discussion.  

Gordan,

The definitive reference for quasi-experimental design is Campbell and Stanley (1971). It has been cited over 3500 times:

Search Google Scholar for "Campbell and Stanley (1971." Click "cited by...." to view more recent references that cite the original work.

Hope this helps.

Don

Hi Gordon! I very much agree with Gina´s comment. I would add that a control group, in addition to a pre-test-post-test design, would further strenghten your design. Due to lack of randomization you cannot infer any causality. And due to a weeker design than the pre-test post-test design with control group, your option is to rule out any confounding effects (see Shadish, Cook, and Campbell). There are strong rationales for using quasi-experiments, but you cannot infer cause-effect. Your job is to argue why a quasi-experiment is better, and why the results are due to collaborative learning.

Lucas: Thank you for your comments.

I agree with you that, from the perspectives of randomization and control, a strict causality is hard to be generated. A strict experiment design should be like this:

Experimental group: R. Pretest   X      Posttest

Control group:           R. Pretest            Posttest

However, one reason for the popularity of quasi-experiment is that oftentimes researchers do not have the luxury of documenting everything due to the time and preparation. In my case described in my original post, I would say that preservice teachers in the control and experiment groups are similar in terms their academic record (GPA), training in the program, the professor who teaches them, the course that I focus on, et al. To a certain degree, I can say the only difference is the use or not use of the collaborative learning pertaining to a very specific aspect about their teaching performance (details are not given here). 

Can I generate causal conclusion? As you said, no. Still, this design is ok. I really cannot go back through the time machine to collect preservice teachers' pretest since they had finished the course. 

----Gina----Your comments are pretty insightful and convincing. I totally agree with you what you said in your post. I will pay close heed to your suggestions in my research. As I mentioned in other posts, I do not have the luxury to go back to collect data by pretest which are exactly corresponding to preservice teachers' specific teaching skills (not to elaborate about the specific aspect about teaching). Still, I can use other information collected prior to preservice teachers' demonstration of their teaching skills. 

With that said, I am delighted to say a little bit about the philosophy of conducting research. As you talked in your post, it is hard to carry out experimental research from a strict sense because social sciences, education for instance, have too many factors to deal with. I is hard, if not impossible, to control other factors while only to focus on those in which you are interested. 

In addition, many theorists and researchers say it is inadvisable to use experimental design in social sciences because findings generated in labs have no any implication in the real world. For example, a researcher finds a teaching method is pretty efficient in terms of increasing students' learning only in the context of controlling their social economic status, intelligence, GPA, and several other factors. The question is: do we have this kind of learning environment in today's classrooms and schools? As a result, researchers are adapting lab environment designs to make them more real and situated. 

But the previous words do not mean that I do not need pretest at all. I only do not have the luxury to have that now. 

I give you my sincere appreciation for your beautiful comments and suggestions. 

Thanks!  I definitely agree about the confound involved in over experimentalizing social science research.  I'd argue that this is equivalently an issue in what many term "hard" sciences as well, but that is another debate entirely.  

In A Quasi- experiential design, you can take the university's cumulative average for the students in the two groups: cohort 2016 vs. cohort 2017.  and compare their performance using the ANCOVA test  In analysis,,for more details see my paper here

http://www.iiste.org/Journals/index.php/JEP/article/view/33057 

Ziad- Thanks!