Good Morning-
I am preparing data for a multilevel path analysis. I am stuck trying to organize my data into levels on the individual and team level. Should I create a separate category to denote the variable as a Level 1 and Level 2 or is that done on the backend (variable view) of SPSS? 

Your help would be greatly appreciated with this matter! 

Cheers,
Dr. Meany 

Below are the details of the variables, levels, and actions required:

Purpose of the Study

The purpose of this study is to explore the positively mediating role of psychological on the relationship between humble leadership and follower creativity, and how knowledge sharing positively moderates psychological safety and creativity in a less high-power distance and less collectivistic culture.

Variable Levels:

Level 1 (Individual Level) -
Gender- G
Age- A
Tenure- T
Creativity (Dependent Variable)- C

Level 2 (Team Level) -
Team Size- TS
Humble Leadership (Independent Variable)- H
Psychological Safety (Mediating Variable)- P
Knowledge Sharing (Moderating Variable)- S

Actions:
1. Depict relationships between team-level variables (Level 2) and individual-level variable (Level 1), through a multilevel analysis-  multilevel path analysis to test this multilevel mediation hypothesis.

2. Additionally a simultaneous multilevel regression procedure within multilevel path analysis to determine if knowledge sharing indeed moderates the relationship between leader humility and follower creativity as mediated by psychological safety

3. A Monte Carlo simulation procedure using open-source software R will test the significance of the influences of mediation and moderated mediation

4. Within-group agreement (depicted as rWG) and reliability (intraclass correlations ICC1 and ICC2) will need to be tested to determine if the aggregation of humble leadership, psychological safety, and knowledge sharing as appropriate as team variables

5. Confirmatory factor analyses (CFAs) to examine discriminant validity.

------------------------------
Brendan Meany
------------------------------

#SPSSStatistics

In my opinion, regardless of the levels of the variables, you can run regression analysis and test your hypotheses. For the following steps, no need to define variables class wise. Keep the data file simple. 
I am replying point wise:
For Action point 1: Test these regressions: 
(i) Humble Relationships (H) --> Creativity (C)
(ii) Psychological Safety (Mediating Variable)- P --> Creativity (C)
(iii) P + H --> C (Since P is a mediating variable, P+H is needed. For moderating, it should be a multiplication, as explained below)
check for the "R2 changed" for (ii) and (iii) along with its "F statistics". which will show you whether the mediating variable has a significant effect on the relationship between H and C

For Action point 2: Run these regressions:
(i) H --> C 
(ii) H*S --> C
Since S is a moderating variable, no direct link between S and C, hence no regression is needed. 
Again, check R squared change and its significance, based on which you will be able to deduce whether Knowledge Sharing really works as a moderating variable.

For Action point 3: Not sure whether Monte Carlo would fit here

For Action point 4: If you find all three (H, P and S) are correlated with each other, then it would be difficult to analyse mediating and moderating effects. In that case, check how to reduce the multicollinearity.

for Action point 5: Not sure whether CFA would fit here. 

Just a curiosity: Are you going to test the whole model in AMOS? If possible, do it. It may show you more meaningful results.


------------------------------
Vandit Desai
------------------------------

Thank you so much for the advice, Vandit. No, I do not have access to AMOS and am unfamiliar with the program. This analysis is for a dissertation so time is of the essence. Originally the student requested that it be completed in Mplus, but that program proved to be much more difficult than anticipated. 

I do agree that leveling would be time consuming and unnecessary for this analysis. 

For the sake of time, I am going to proceed in SPSS. I appreciate your time and help! 

Cheers,
Dr. Meany

------------------------------
Brendan Meany
------------------------------

Original Message:
Sent: Tue August 24, 2021 05:39 AM
From: Vandit Desai
Subject: Organizing Data for Multilevel Path Analysis

In my opinion, regardless of the levels of the variables, you can run regression analysis and test your hypotheses. For the following steps, no need to define variables class wise. Keep the data file simple. 
I am replying point wise:
For Action point 1: Test these regressions: 
(i) Humble Relationships (H) --> Creativity (C)
(ii) Psychological Safety (Mediating Variable)- P --> Creativity (C)
(iii) P + H --> C (Since P is a mediating variable, P+H is needed. For moderating, it should be a multiplication, as explained below)
check for the "R2 changed" for (ii) and (iii) along with its "F statistics". which will show you whether the mediating variable has a significant effect on the relationship between H and C

For Action point 2: Run these regressions:
(i) H --> C 
(ii) H*S --> C
Since S is a moderating variable, no direct link between S and C, hence no regression is needed. 
Again, check R squared change and its significance, based on which you will be able to deduce whether Knowledge Sharing really works as a moderating variable.

For Action point 3: Not sure whether Monte Carlo would fit here

For Action point 4: If you find all three (H, P and S) are correlated with each other, then it would be difficult to analyse mediating and moderating effects. In that case, check how to reduce the multicollinearity.

for Action point 5: Not sure whether CFA would fit here. 

Just a curiosity: Are you going to test the whole model in AMOS? If possible, do it. It may show you more meaningful results.


------------------------------
Vandit Desai
------------------------------

One last question, before adding P+H and multiplying HxS, should I center the variables? I have seen videos on Youtube where the moderation between variables was centered (ZScore) before multiplying them together. 

Cheers,
Dr. Meany

------------------------------
Brendan Meany
------------------------------

Original Message:
Sent: Tue August 24, 2021 05:39 AM
From: Vandit Desai
Subject: Organizing Data for Multilevel Path Analysis

In my opinion, regardless of the levels of the variables, you can run regression analysis and test your hypotheses. For the following steps, no need to define variables class wise. Keep the data file simple. 
I am replying point wise:
For Action point 1: Test these regressions: 
(i) Humble Relationships (H) --> Creativity (C)
(ii) Psychological Safety (Mediating Variable)- P --> Creativity (C)
(iii) P + H --> C (Since P is a mediating variable, P+H is needed. For moderating, it should be a multiplication, as explained below)
check for the "R2 changed" for (ii) and (iii) along with its "F statistics". which will show you whether the mediating variable has a significant effect on the relationship between H and C

For Action point 2: Run these regressions:
(i) H --> C 
(ii) H*S --> C
Since S is a moderating variable, no direct link between S and C, hence no regression is needed. 
Again, check R squared change and its significance, based on which you will be able to deduce whether Knowledge Sharing really works as a moderating variable.

For Action point 3: Not sure whether Monte Carlo would fit here

For Action point 4: If you find all three (H, P and S) are correlated with each other, then it would be difficult to analyse mediating and moderating effects. In that case, check how to reduce the multicollinearity.

for Action point 5: Not sure whether CFA would fit here. 

Just a curiosity: Are you going to test the whole model in AMOS? If possible, do it. It may show you more meaningful results.


------------------------------
Vandit Desai
------------------------------

OK, check first the correlations among H, P, and S 
(Action point no. 4 should be the first step in your analysis)

If the correlation matrix suggests high collinearity among them (Whether among all three of them or between any two of them), then the issue of multi-collinearity exists. 
In that case (when there is multi-collinearity), go with centering the variables before running the regressions. 

If multi-collinearity is not present (that is, none of them is significantly correlated with any other of them), then centering is not required.

------------------------------
Vandit Desai
------------------------------

Original Message:
Sent: Wed August 25, 2021 02:50 PM
From: Brendan Meany
Subject: Organizing Data for Multilevel Path Analysis

One last question, before adding P+H and multiplying HxS, should I center the variables? I have seen videos on Youtube where the moderation between variables was centered (ZScore) before multiplying them together. 

Cheers,
Dr. Meany

------------------------------
Brendan Meany
------------------------------


Original Message:
Sent: Tue August 24, 2021 05:39 AM
From: Vandit Desai
Subject: Organizing Data for Multilevel Path Analysis

In my opinion, regardless of the levels of the variables, you can run regression analysis and test your hypotheses. For the following steps, no need to define variables class wise. Keep the data file simple. 
I am replying point wise:
For Action point 1: Test these regressions: 
(i) Humble Relationships (H) --> Creativity (C)
(ii) Psychological Safety (Mediating Variable)- P --> Creativity (C)
(iii) P + H --> C (Since P is a mediating variable, P+H is needed. For moderating, it should be a multiplication, as explained below)
check for the "R2 changed" for (ii) and (iii) along with its "F statistics". which will show you whether the mediating variable has a significant effect on the relationship between H and C

For Action point 2: Run these regressions:
(i) H --> C 
(ii) H*S --> C
Since S is a moderating variable, no direct link between S and C, hence no regression is needed. 
Again, check R squared change and its significance, based on which you will be able to deduce whether Knowledge Sharing really works as a moderating variable.

For Action point 3: Not sure whether Monte Carlo would fit here

For Action point 4: If you find all three (H, P and S) are correlated with each other, then it would be difficult to analyse mediating and moderating effects. In that case, check how to reduce the multicollinearity.

for Action point 5: Not sure whether CFA would fit here. 

Just a curiosity: Are you going to test the whole model in AMOS? If possible, do it. It may show you more meaningful results.


------------------------------
Vandit Desai

Original Message:
Sent: 8/26/2021 3:00:00 AM
From: Vandit Desai
Subject: RE: Organizing Data for Multilevel Path Analysis

OK, check first the correlations among H, P, and S 
(Action point no. 4 should be the first step in your analysis)

If the correlation matrix suggests high collinearity among them (Whether among all three of them or between any two of them), then the issue of multi-collinearity exists. 
In that case (when there is multi-collinearity), go with centering the variables before running the regressions. 

If multi-collinearity is not present (that is, none of them is significantly correlated with any other of them), then centering is not required.

------------------------------
Vandit Desai
------------------------------


Original Message:
Sent: Wed August 25, 2021 02:50 PM
From: Brendan Meany
Subject: Organizing Data for Multilevel Path Analysis

One last question, before adding P+H and multiplying HxS, should I center the variables? I have seen videos on Youtube where the moderation between variables was centered (ZScore) before multiplying them together. 

Cheers,
Dr. Meany

------------------------------
Brendan Meany
------------------------------

Oh. thanks John for the clarification. I had read it somewhere but did not cross-check. 
Dr. Meany, in this case, please ignore my last post. ​

------------------------------
Vandit Desai
------------------------------

Original Message:
Sent: Thu August 26, 2021 08:41 AM
From: Jon Peck
Subject: Organizing Data for Multilevel Path Analysis

Centering has nothing to do with multicollinearity.  While it may help in interpreting interaction terms, the statistical results are equivalent.

--

Jon K Peck
jkpeck@gmail.com

Original Message:
Sent: 8/26/2021 3:00:00 AM
From: Vandit Desai
Subject: RE: Organizing Data for Multilevel Path Analysis

OK, check first the correlations among H, P, and S 
(Action point no. 4 should be the first step in your analysis)

If the correlation matrix suggests high collinearity among them (Whether among all three of them or between any two of them), then the issue of multi-collinearity exists. 
In that case (when there is multi-collinearity), go with centering the variables before running the regressions. 

If multi-collinearity is not present (that is, none of them is significantly correlated with any other of them), then centering is not required.

------------------------------
Vandit Desai
------------------------------


Original Message:
Sent: Wed August 25, 2021 02:50 PM
From: Brendan Meany
Subject: Organizing Data for Multilevel Path Analysis

One last question, before adding P+H and multiplying HxS, should I center the variables? I have seen videos on Youtube where the moderation between variables was centered (ZScore) before multiplying them together. 

Cheers,
Dr. Meany

------------------------------
Brendan Meany


Original Message:
Sent: Tue August 24, 2021 05:39 AM
From: Vandit Desai
Subject: Organizing Data for Multilevel Path Analysis

In my opinion, regardless of the levels of the variables, you can run regression analysis and test your hypotheses. For the following steps, no need to define variables class wise. Keep the data file simple. 
I am replying point wise:
For Action point 1: Test these regressions: 
(i) Humble Relationships (H) --> Creativity (C)
(ii) Psychological Safety (Mediating Variable)- P --> Creativity (C)
(iii) P + H --> C (Since P is a mediating variable, P+H is needed. For moderating, it should be a multiplication, as explained below)
check for the "R2 changed" for (ii) and (iii) along with its "F statistics". which will show you whether the mediating variable has a significant effect on the relationship between H and C

For Action point 2: Run these regressions:
(i) H --> C 
(ii) H*S --> C
Since S is a moderating variable, no direct link between S and C, hence no regression is needed. 
Again, check R squared change and its significance, based on which you will be able to deduce whether Knowledge Sharing really works as a moderating variable.

For Action point 3: Not sure whether Monte Carlo would fit here

For Action point 4: If you find all three (H, P and S) are correlated with each other, then it would be difficult to analyse mediating and moderating effects. In that case, check how to reduce the multicollinearity.

for Action point 5: Not sure whether CFA would fit here. 

Just a curiosity: Are you going to test the whole model in AMOS? If possible, do it. It may show you more meaningful results.


------------------------------
Vandit Desai

No worries at all. I appreciate the help from both you and Jon. This message board is such a cool concept. Assisting strangers with these challenging tasks is very collegial. I will certainly pay it forward! 

Cheers,
Dr. Meany

------------------------------
Brendan Meany
------------------------------

Original Message:
Sent: Thu August 26, 2021 03:26 PM
From: Vandit Desai
Subject: Organizing Data for Multilevel Path Analysis

Oh. thanks John for the clarification. I had read it somewhere but did not cross-check. 
Dr. Meany, in this case, please ignore my last post. ​

------------------------------
Vandit Desai
------------------------------


Original Message:
Sent: Thu August 26, 2021 08:41 AM
From: Jon Peck
Subject: Organizing Data for Multilevel Path Analysis

Centering has nothing to do with multicollinearity.  While it may help in interpreting interaction terms, the statistical results are equivalent.

--

Jon K Peck
jkpeck@gmail.com

Original Message:
Sent: 8/26/2021 3:00:00 AM
From: Vandit Desai
Subject: RE: Organizing Data for Multilevel Path Analysis

OK, check first the correlations among H, P, and S 
(Action point no. 4 should be the first step in your analysis)

If the correlation matrix suggests high collinearity among them (Whether among all three of them or between any two of them), then the issue of multi-collinearity exists. 
In that case (when there is multi-collinearity), go with centering the variables before running the regressions. 

If multi-collinearity is not present (that is, none of them is significantly correlated with any other of them), then centering is not required.

------------------------------
Vandit Desai


Original Message:
Sent: Wed August 25, 2021 02:50 PM
From: Brendan Meany
Subject: Organizing Data for Multilevel Path Analysis

One last question, before adding P+H and multiplying HxS, should I center the variables? I have seen videos on Youtube where the moderation between variables was centered (ZScore) before multiplying them together. 

Cheers,
Dr. Meany

------------------------------
Brendan Meany


Original Message:
Sent: Tue August 24, 2021 05:39 AM
From: Vandit Desai
Subject: Organizing Data for Multilevel Path Analysis

In my opinion, regardless of the levels of the variables, you can run regression analysis and test your hypotheses. For the following steps, no need to define variables class wise. Keep the data file simple. 
I am replying point wise:
For Action point 1: Test these regressions: 
(i) Humble Relationships (H) --> Creativity (C)
(ii) Psychological Safety (Mediating Variable)- P --> Creativity (C)
(iii) P + H --> C (Since P is a mediating variable, P+H is needed. For moderating, it should be a multiplication, as explained below)
check for the "R2 changed" for (ii) and (iii) along with its "F statistics". which will show you whether the mediating variable has a significant effect on the relationship between H and C

For Action point 2: Run these regressions:
(i) H --> C 
(ii) H*S --> C
Since S is a moderating variable, no direct link between S and C, hence no regression is needed. 
Again, check R squared change and its significance, based on which you will be able to deduce whether Knowledge Sharing really works as a moderating variable.

For Action point 3: Not sure whether Monte Carlo would fit here

For Action point 4: If you find all three (H, P and S) are correlated with each other, then it would be difficult to analyse mediating and moderating effects. In that case, check how to reduce the multicollinearity.

for Action point 5: Not sure whether CFA would fit here. 

Just a curiosity: Are you going to test the whole model in AMOS? If possible, do it. It may show you more meaningful results.


------------------------------
Vandit Desai