Sample Size (6 LV each one is linked to another one)

Frequently asked questions about PLS path modeling.
Post Reply
Julen
PLS User
Posts: 17
Joined: Wed Apr 24, 2013 7:38 am
Real name and title:

Sample Size (6 LV each one is linked to another one)

Post by Julen » Mon May 27, 2013 4:46 pm

Dear all,

I need to calculate my sample size for smartPLS. I have 6 LV (formative) and each one is linked to another LV (this would be the 7th)called "performance" (It might be reflective, I have drawn it like formative, though). There are no other links between the 6 LV. I would like to include a moderating effect, I guess I could divide the sample in 2 (pionner businesses and followers), run the PLS twice and see if the results are significantly different. I wanna see which LV has the greater impact in Performance for each group (pionners and Followers), which one the less impact...

But continuing with the simple size, I guess I have these options:

1) I have read that I could use POWER Analysis:

How many arrows arrive in each LV?
Choose the biggest number
Use this number as number of predictors in the multiple regression procedure in the G*Power to have the sample size

I have the program but don´t not well how it Works, with a power of 0,8, and an effect size of 0,15 (see cohen 92 table 1 for regressions), I´m not sure which tests do I need. Does it depend on my hiphotesys? For example for a t test it says I need only a sample of 43

2) 10 (or 5) TIMES RULE

Considering that no LV has more than 6 indicators, In my case it would mean the maxiumn number of arrows pointing a LV, i.e., 6 x 10= 60 minimun, isn´t it?

3) Cohen 92 --> in table 2 --> 6 regressions, medium effect, alpha 0,05--> sample of 97

DOES IT LOOK RIGHT? WHICH ONE SHOULD I USE?
(thank you very much in advance)
Last edited by Julen on Wed Jun 05, 2013 12:50 pm, edited 1 time in total.

User avatar
Hengkov
PLS Super-Expert
Posts: 1618
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov » Sun Jun 02, 2013 1:44 pm

Hi,
You must have suciffient sample (consistent at large). 60 is minimum, 100 it's OK for support power analysis.

Regards,

Julen
PLS User
Posts: 17
Joined: Wed Apr 24, 2013 7:38 am
Real name and title:

Post by Julen » Mon Jun 03, 2013 3:08 pm

Thanks very much, Mr Hengkov.

I have talked to an expert, and I was wrong, I am NOT including a moderating effect but the thing I wanna do is a GROUP COMPARISON (in Henseler 2009 et al. it´s explained, 3.5.).

That means I have to divide my sample into two and then do an specific analysis with the Beta coefficients, so I need the minum sample but FOR EACH group, that means that if the minumun sample is 6.10=60 ( 6 is the maximun number of arrows arriving at a variable), I would need it for each group, that means minumun 120.

Let´s say I get 150 responses. Would it be enough for supporting the power analysis? or would I need around 200? I don´t really know how important this power analysis is...

THANK YOU!

User avatar
Hengkov
PLS Super-Expert
Posts: 1618
Joined: Sun Apr 24, 2011 10:13 am
Real name and title: Hengky Latan
Location: AMQ, Indonesia
Contact:

Post by Hengkov » Tue Jun 04, 2013 10:43 am

Hello,

You have moderating categorical variable?
Yes, you can ran PLS-MGA. Many approach required sample size each group not different. I think 120 is enough with each group 60. You must test measurement invariance too.

Regards,

Julen
PLS User
Posts: 17
Joined: Wed Apr 24, 2013 7:38 am
Real name and title:

Post by Julen » Tue Jun 04, 2013 11:09 am

Hi Mr. Hengkov and others,

I don´t know exactly how is it called, I don´t think it´s a moderating effect, I just wanna test that the effect of each of the six latent variables in the other (that would be the seventh, it´s called performance) latent variable is different depending on the firms being pioneers or followers.

Yes, I´ve been told that it´s a GROUP COMPARISON (PLS-MGA).

I would divide the simple into 2 groups (I have some Likert scale questions). If you tell me that sometimes both samples have to have the same number of responses, I guess I would use "the median Split" criteria to divide the sample into 2 (not to loose any responses), unless I can find lots of responses (around 200).

OK, I´ll keep in mind the measurement invariance, don´t know yet how it Works or what else do I need to test, but I´ll think about it.

The first thing I wanted to make sure about is that I would be able to get the minimun sample size. Is there anything else important (I mean, regarding to PLS...) that we have to keep in mind before doing the field research? I would like to be sure that the money I pay for getting the responses is well invested...

THANK YOU VERY MUCH, this forum is a great help.

Post Reply