According to my knowledge the sample size of approximately 300 is good enough as i found few papers with this sample size ..but its good if u confirm it from yours supervisor   

Dear Tricia

The sample size decisions are subject to the diversity in population on the variables of research and desired accuracy of results. However, the well accepted thumb rule for sample size for studies using CFA and SEM for data analysis is a minimum of number of measurement variables (statements in your questionnaire in survey based research) * 10. That means, if you are using a single variable in your study having 20 statements; the minimum sample size you should take will be 20*10= 200. The minimum sample size for for studies using CFA and SEM  should not be below 200 in any case. For multi- variable studies the sample size is worked out using the variable having maximum statements.

If you are truly conducting an experimental study then you should ask yourself whether you should use SEM to tackle your research hypotheses. Generally, the purpose of experimental studies is to assess causal relationships and in particular, examine and conclude on expected/hypothesized interaction effects. SEM is not the right tool for such investigations.

Thank you everyone these are all very useful responses. I find your responses more pointed than anything I've seen online so this is truly very helpful. Is there an article that you can point me to for reference? Again, thanks so much.

Edward, thanks for your response also. Can you please elaborate on what disqualifies SEM when doing experiments or point me to an article explaining same. Thanks much.

Correlational analyses (e.g., regression or SEM) based on single cross-sectional survey data cannot evidence causality. This is because for A to cause an effect B, A must take place before B and so, you will need a reasonable amount of time to separate the measures of A and subsequently B. To truly examine and evidence causal relationships, you will need to conduct longitudinal studies or experimentation. Even with longitudinal data, SEM poses additional difficulties in that modelling two- or three- way interactions are very problematic in practice in SEM. But to answer your question better regarding the preference for experimentation, please see http://www.uk.sagepub.com/upm-data/23639_Chapter_5___Causation_and_Experimental_Design.pdf or some other research method text.

The minimum in terms of sample depends on some search parameters : knowledge or not the population ( N) ; research problem ; number of variables and theory on the subject of research. But at first, when it is not known data on the population (N ) for each variable or parameter , in general , must have at least 10 samples. Then multiply each variable * 10. Some authors comment that, as Hair (2013 ) .

Hair Jr., J. F .; CELSI , M. W .; ORTINAU , D. J .; and Bush , R. P. ( 2013) . Essentials of marketing research . (3 th . Ed.). New York: McGraw-Hill .

Please also note that the requirements for sample size are much much smaller if you use Partial Least Squares Modeling (PLS). PLS does the same thing than SEM.

According Hair et al., 2001, The discussion of statistical power shows that the substantial impact of sample size operating in the attainment of statistical significance, in large or small sample sizes. Whenever the sample sizes exceeding 200 or 440 respondents, the researcher should examine all results significativospara sure they have practical significance due to increased statistical power due to the sample size.

This is a quotation from Haenlain and Kaplan (2004):  A Beginner’s Guide to Partial Least Squares Analysis. From this you ca see that there is a big difference between SEM and PLS as far as sample size requirement is concerned.

For covariance-based SEM, it is generally advisable that the “sample size should exceed 100 observations regardless of other data characteristics to avoid problematic solutions and obtain acceptable fit concurrently” (Nasser&Wisenbaker,2003, p. 754), and many researchers even recommend a minimum sample size of 200 cases (e.g., Marsh et al., 1998) to avoid results that cannot be interpreted, such as negative variance estimates (i.e., Heywood cases) or correlations greater than one (i.e., improper solutions; Dillon, Kumar, & Mulani, 1987). PLS, on the other hand, is applicable even under conditions of very small sample sizes. Although a detailed discussion of this point can only bemadefor each individual model based on statistical power analysis, a Monte Carlo simulation performed by Chin and Newsted (1999) indicated thatPLScan be performed with a sample size as low as 50, and H. Wold even “analysed 27 variables using two latent constructs with a data set consisting of ten cases” (Chin, Marcolin,& Newsted, 2003a, Appendix A, p. 5).

Matti Haverila thank you so much this is most helpful. I had been trying to find good references for PLS and so I am very grateful for your references here.

I have more if you need! I have been using the SmartPLS software for a few years and find it very useful. Wynne W. Chin is another good source in PLS.

As a first, you should decide to use whether CB-SEM or PL-SEM. Then you can follow the regulations and rules for sample size.

Dear Tricia,

By seeing the question, I was very interested to write an answer but when saw the answers, there is nothing left. All the experts have written and explained in very very simple way.

Specially comments by Dr. SS Bhakar are self-explanatory.

I request to all, that if you like answers then please hit up-votes  and increase mutual RG Score.

Tricia

Try these papers

What is the minimum sample size for multiple-group path analysis?

The rules are the same whether you have multi-group or single groups analysis.

What is the difference between sample representativeness and the right sample size for adequate statistical power ?

Good question and informative responses. Thanks!

There is another dimension to this question and that is related to the use of MGA (multi-group analysis) within PLS. As you may know when comparing the mean values of two groups, the bigger the sample size, even quite small differences in the mean values can be significantly different from each other. This same logic is true in MGA-PLS. There is no rule of thumb, but if the sample size is in thousands, then a mean value difference of 0.1 or so can be significantly different. From managerial point of view, this is is not meaningful of course. If the sample size is in tens on the other hand, a big difference in mean values (maybe about 0.8 or so) is needed, which from academic point if view might bring an uninteresting result. So I guess having a sample size in hundreds should be the target. This of course is quite a bit more than the minimum sample size guidelines usually presented for PLS.

Interesting piece of information...thank you for sharing it.

Edward, your first answer is a myth, nothing else. Your second answer ignores structured mean models. At least please justify your first answer.

Using G power software can help determining sample size

@Mitra. Thank you so much. I have never heard of G power software.

Hair et al. describes nicely about the sample size in SEM. Lots of issues are involved in sample size calculation: distribution, no. of latent variables, communalities of latent factors, having fewer than three measured items in each construct etc. It should be better to take adequate sample based on that particular situation rather strict to the rules. Large sample is always in favor of good outcome.

It depends on how diverse your population and how many latent variable but in average you need between 80 - 150 samples. This number is required to achieve stable result.

G*Power is an open source program for power analysis and sample size calculations. There are also quite few tutorials and a manual, so it will help you determine your sample size according to the analysis you are planning and the expected standard error.

It's not wise to depend on software only to calculate the required sample size. It's better to judge the context of the study and the basic rules for GOF. The larger sample will produce the better outcome no doubt.

Sample size isn't necessarily a maximization problem, Muhammed. This is just as much a myth as the claim that SEM is not suitable for experimental research designs.

You are absolutely right, Arnold. I never mentioned about the limitations of SEM in experimental study. Application of SEM will be more fruitful when it will be calculated based on a theoretical framework. The larger sample will be more reliable to produce valid SEM model in social research.

Christian's my first name. Not important. The myth about SEM and experimental designs is at the top (not from you). No, maximizing the sample size doesn't necessarily lead to better results (example): https://onlinelibrary.wiley.com/doi/abs/10.1002/asi.23212

Read my post about What is a good sample size for Structural Equation Modeling (SEM) here: www.sarpublisher.com/what-is-a-good-sample-size-for-structural-equation-modeling-sem/

Hi, I need to know how to use gpower for a route analysis, What options do I have to select to get the power of the model?

Also, I have to keep in mind that the variables are binary and that all the variables in the model are observable.

Thanks.

Kline (1998) and field (2009) states that an adequate sample size performing a SEM or Path analysis is equat to 10 times the amount of the parameters considered in th study, but the best sample is 20 times.

@ Mohd Fuad Salleh please do you have citation for the figures you mentioned ?

Soufiane Boukarta can you please share reference regarding best sample being 20 times the number of questions?

I have mentioned the reference and here below I am giving you the detail of the reference:

Kline, Rex B. (1998). Principles and practice of structural equation modeling.

NY: Guilford Press.

Thanks Soufiane Boukarta

10*of the constructs

Dear Mohd Fuad Salleh Could you please provide a reference that I can use for the sample size (80-150)?

Thanks a lot