## Procedere in case of a low sample size (too low?)

Alex_
PLS Junior User
Posts: 1
Joined: Thu Oct 15, 2020 3:37 pm
Real name and title: Alexander Brodsky (doctoral candidate)

### Procedere in case of a low sample size (too low?)

Dear Community,

as a part of my dissertation, I try to find a solution to analyze the data from our research project by SEM. During my methodological research, I came across SmartPLS and read about PLS-SEM.
I did some research regarding the minimum sample (I know, it is a regularly discussed topic ;-)) and read about different guidelines (10 time-rule; R-squared method, methods by Kock & Hadaya, 2016).

Some information about the proposed model (attached):
We collected pre-, process- and post-data from students doing an internship. In one part of my analysis, I want to examine whether (a) the expectations (towards the development of different competence facets; latent constructs) exert an influence on the activities the students are engaged with during the internship and (b) whether the activities at the workplace exert an influence on the perceived competence development.

The expectations and the perceived competence development use the same contents and each of the scales include between 5 and 11 (!!!) items (latent constructs). The activities are included in form of the averages of five measurements during the internship (scale between 1 = never and 4 = often).

Problem: Regardless of the criteria of the minimum sample size which I use (e.g. in case the time rule, I needed 110 students due to the scale with 11 items), I don´t have the “minimum sample” (and cannot repeat or recollect the data due to limitations of the project). My sample size is around 70 students.

My questions are:
- What are main statistical problems of analyses with not "enough" people? (as the number of the students in enrolled in the study program examined is not that big, the representation of the sample size wouldn´t be a big problem)
- Is it possible to carry out and interpret the analysis despite the problem (or are they useless through problems that have arisen)? How could I argue to perform the analysis anyway? (any literature on that?)
- More general: What´s the difference between using latent variables (and indicate which manifest variables address a latent construct) and using averages of the scales to conduct the SEM? (the results differentiate if I compare these two different ways)

I would be very happy about a feedback, because I did not find any hints for this specific question

Best
Alex
Attachments
Untersuchungsdesign.pptx [Automatisch gespeichert].jpg (76.58 KiB) Viewed 3378 times