In SEM, a model fit between one IV (X) and two DVs (Y1 & Y2) is adequate (RMSEA=0.08; CFI=0.92; GFI=0.90 & NFI=0.86) with N=134.
The causality between X and Y1 & Y2 are found to be significant at 5% significant level.
However, the standard beta values between X and Y1 and X and Y2 are low (i.e. -0.28 and +0.32) respectively.
How to report such findings for social science publications when literature shows a greater strength of such relationships (in respective directions).
I think,the strength of values are relative term, as we always refer to the effect size of previous studies. If the previous are studies are having same effect size , then there is no problem as your study is aligned with literature. However, significantly greater or lesser effect size than literature requires possible reasons to be explained by the researchers ( may be culture, context or variable size or respondents etc).
Regards
Rafique
Thank you so much Rafique for your reply..
Actually, in the study the impact is lesser than what is usually mentioned in the literature. However, I will take cultural aspects into account for further explanation.
I agree with Muhammad, you may need to look at the variable size
The strength of causality and its significance are related.
First of all, you need to report how much the changes in DV for evry unit change in IV. And secondly, you report whether the said changes is statistically significant or not significant.
I would agree that strength of causality varies by culture depending on your IV and DVs. You also have a control variable to think about (typical ones are gender but you will have ones for your context). Your standardised values may change significantly if controls are added in or if controls are not important then they may not. I would expect that many of the papers you are comparing your values to include such controls in their methods and in their analysis so therefore you may not be comparing like with like. On a final note be careful in comparing your relatively straightforward model with the coefficients found by other authors in a larger published model as they are not truly comparable. Finally when you say DV and IV, it sounds to me like it is path analysis ... if it is SEM then perhaps you need to ensure that you are speaking about constructs or latent variables or some such terminology ...
Best of luck with your research
The causality between latent variables (Independent) and Observed variables (Dependent) can vary . You need to explain why there is a low relationship. Sometimes the constructs might be having relatively low importance to the group in which you conducted the study. Compare the context , sample size and constructs taken in the existing literature with your study context and then you will get reasons to justify the low relationship reported in your study.
Dear Sayoni,
can I suggest that you are careful about the terminology that you use. SEM and path analysis will produce estimates of the strength of the (usually) linear relationship between variables (be they latent or observed), but it's only under certain conditions that these estimates be described as a causal relationship. Usually we have to have elements of an experiment in place (random allocation and manipulation of the IV) before we can make any inferences about causality. Otherwise we can just talk about associations.
The question of why your estimates may be different (lower) than other research findings is a different question. As other have noted it may be due to the scaling of your IV's (e.g. scale time in days and you'll get a much smaller effect than using years), measures being less reliable in your sample (low reliability will attenuate effects), or maybe some other characteristics of your sample (e.g. the association between age and height will be high for a sample of children aged between 0 and 10 years, but very low or maybe even negative for a sample of people aged 60 to 90 years). Also have a look at where your estimates are in relation to the confidence intervals of previously reported estimates.
best wishes
mark
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling