Multiple regression: Insignificant ANOVA (model2), significant individual variable. What to do now?

[falcofly]

Online_question_output.spv

Dear all,

I have a question with regards to my dissertation.
I doing research on the effects of (1) social network resources on (3) employability, with a mediation variable (2) of informal learning. These are all latent variables/constructs made up of several scales/underlying variables.
I'm using gender, age and tenure as control variables.

At the moment I'm stuck with a hierarchical multiple regression interpretation of Feedback Exchange and Information Seeking (both scales underlying Informal Learning) on Personal Flexibility (one of the underlying scales of Employability)

Hence, to clarify, in this regression I'm treating the mediation variables as Independent Variables/predictors.

Now, the problem that I have is illustrated in the output-document that I uploaded with this message:
My R2 F Significance is significant, but my complete model shown in the ANOVA output under model 2 is not significant. Additionally, the unique variation of variable informal seeking is again significant.

Model 1 R2= 0.002
Model 2 R2= 0.069                            Sig F Change: 0.019 (significant)

ANOVA output:
Model 2:                                          0.143 (not significant)

Coefficients:
Info_Seeking: Beta (.305),                  Sig 0.006

Moreover, if I take out all the control variables and variable feedback exchange, I get a simple linear regression in which everything is significant again.

My question:  What am I supposed to report here?

Do I just say that the overall model is not significant, and therefore nothing can be argued. (and don't continue reporting the other values)
Or do I mention the overall model is not significant due to the insignificance of the control variables (the correlation matrix shows none of the other variables except Information Seeking is correlating with the outcome/dependent variable Personal Flexibility) and feedback exchange, and say the information seeking alone is in fact significant?

As one can see I do not have troubles with multicorrelation or anything.

Looking very much forward to your advice.
It would help a lot!

Best,

Falco

[Art Kendall]

Do you have enough cases to detect an effect?

Did your visualizations, e.g., parallel coordinate plots, 3 D scatter plots, etc. show any linear fits?
Did you try fitting linear and loess fits to  your scatterplots?

Are the latent variables made up of scales or of items?
how did you actually measure personal flexibility, age,  tenure, info seeking, FB_Exchange? What construct do these names stand for?

Did you double enter or proofread your data?

I do not see variable labels. Did you check that your variables view was complete before you ran your data?

Who are your respondents?
On 4/13/2013 5:03 PM, falcofly [via SPSSX Discussion] wrote:

Online_question_output.spv Dear all, I have a question with regards to my dissertation. I doing research on the effects of (1) social network resources on (3) employability, with a mediation variable (2) of informal learning. These are all latent variables/constructs made up of several scales/underlying variables. At the moment I'm stuck with a multiple regression interpretation of Feedback Exchange and Information Seeking (both scales underlying Informal Learning) on Personal Flexibility (one of the underlying scales of Employability) Hence, to clarify, in this regression I'm treating the mediation variables as Independent Variables/predictors. Now, the problem that I have is illustrated in the output-document that I uploaded with this message: My R2 F Significance is significant, but my complete model shown in the ANOVA output under model 2 is not significant. Additionally, the unique variation of variable informal seeking is again significant. Model 1 R2= 0.002 Model 2 R2= 0.069 Sig F Change: 0.019 (significant) ANOVA output: Model 2: 0.143 (not significant) Coefficients: Info_Seeking: Beta (.305), Sig 0.006 Moreover, if I take out all the control variables and variable feedback exchange, I get a simple linear regression in which everything is significant again. My question: What am I supposed to report here? Do I just say that the overall model is not significant, and therefore nothing can be argued. (and don't continue reporting the other values) Or do I mention the overall model is not significant due to the insignificance of the control variables (the correlation matrix shows none of the other variables except Information Seeking is correlating with the outcome/dependent variable Personal Flexibility) and feedback exchange, and say the information seeking alone is in fact significant? As one can see I do not have troubles with multicorrelation or anything. Looking very much forward to your advice. It would help a lot! Best, Falco

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/Multiple-regression-Insignificant-ANOVA-model2-significant-individual-variable-What-to-do-now-tp5719459.html

To start a new topic under SPSSX Discussion, email [hidden email]
To unsubscribe from SPSSX Discussion, click here.
NAML

[falcofly]

I'll answer all question subsequently:

1. Yes, I have enough cases (120 respondents) with 5 Independent Variables being measured that should be enough. Moreover, other regressions before this, were significant.
 This is actually the only regression of which the model is insignificant.

2. I don't exactly know what you mean by linear fits, but my scatterplot and P-P plots show a normal distribution in that all assumptions for a regression analysis are met. The individual variables were all roughly normal, and with enough cases etc.

3. The latent variables are made up of both items and scales. All variables were measured on Likert-scales ranging from 1-5, except employability, which was measured on a scale from 1-6 (totally disagree-totally agree) Next, I performed a Maximum Likelihood Factor Analysis to group the items into meaningful factors. As explained, Informal Learning had two factors/subscales (Feedback Exchange and Information Seeking) based on Kyndt, Dochy and Nijs (2009) "Learning conditions for non-formal and informal workplace learning". and employability showed three factors/scales -Occupational Expertise, Anticipation and Optimization, and Personal Flexibility, based on Van der Heijde & Van der Heijden (2006) "A competence-based and multidimensional operationalization and measurement of employability".
All scales were reliable based on a high Cronbach alpha.

Age was measured on a continuous scale asking how old people were "1-100", tenure was asked similarly based on "how many years have you worked in this function", and gender speaks for itself.(dichotomous)

4. I'm not sure what you mean by proofread, but I do know that my data is sound as it has been used before for other research.

5. I don't understand what you mean by completing the variable view. All variables are accounted for: "FB_exchange, Info_seeking, Age, Tenure, Gender, Personal Flexibility".

6. The respondents are employees of a 911 Rescue service/ambulance agengy, with a size of 120 respondents.

Does this answer your questions?