Hi, I planned to develop a formative model for an index, as the items of each construct are not literally or logically intercorrelated. However, after running the Confirmatory Tetrad Analysis (CTA), the results suggest that my model should be reflective, as most of the items are not significant. It seems I may need to develop a scale instead. Is passing the CTA so crucial that without it, I cannot claim this to be a formative model? Or this is only needed if I need to develop a scale, so I need to confirm this is reflective first?
The choice of formative vs reflective should be determined primarily by the design of the set of measures. If they are not intentionally correlated, then they are probably formative.
Dear James, I ran a CTA Analysis and found that 4 of 9 cases have p-value less than 0.05, meaning 44% of all combinations and led to a formative model. Is this enough to say that the model had to be formative? I'm asking you because If I am looking at the CI low adj and CI up adj most of the correlations are -+, meaning that the model has to be reflective. Thank you very much!
Usually we lean towards a reflective model whenever reasonably possible. So, if the reliability scores and AVE are good enough, then I wouldn't even rely on the CTA. However, if you're trying to justify modeling as formative, then the CTA can provide some support. In general, it is simply easier to validate reflective models.
Thanks James! 1. You changed the arrows direction to formative type of the 3 dimensions of Burnout from Management, However, the arrows of the 6 items in the same Burnout from Management variable are still in reflective form. Could you explain the reason? 2. How is the process of determining the type of measurement model of the second-order latent variable construct? especially if we have multidimensional constructs.
1. When using the repeated indicator approach in SmartPLS for higher-order constructs that are formative, the repeated indicators should be modeled reflectively (arrows pointing at the repeated indicators). This is because the repeated indicator approach is inherently reflective in nature. (See Wetzels, M., Odekerken-Schröder, G., & Van Oppen, C. (2009). Using PLS path modeling for assessing hierarchical construct models: Guidelines and empirical illustration. MIS Quarterly, 33(1), 177-195. OR Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2017). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM). Sage Publications.) 2. The best approach is conceptual. Consider the first order dimensions and whether they are required to be correlated in a rational context. For example, is it possible for emotional exhaustion to increase while depersonalization increases? If so, then this is probably formative. If not, then this is probably reflective.
Hi James, thank you very much for your videos. after you decided unsupportive boss should be formative then the arrows were inverted for indicators not for the constructs. but for burnout management instead of arrows of indicators, arrows of constructs were inverted. could you pls explain, why?
Buenas tardes Dr. James, excelente explicación, tengo una consulta: El análisis de tétrada confirmatoria se puede aplicar para una variable de segundo orden reflexiva-reflexiva?
Creo que habría que hacerlo con las puntuaciones de las variables latentes para las dimensiones de primer orden, en lugar de con los indicadores repetidos.
Yes, report it how it is, but also check the CR value. If it is above 0.700 then this might be sufficient evidence of convergent validity. AVE is a strict measure of convergent validity. Malhotra and Dash (2011) note that "AVE is a more conservative measure than CR. On the basis of CR alone, the researcher may conclude that the convergent validity of the construct is adequate, even though more than 50% of the variance is due to error.” (Malhotra and Dash, 2011, p.702). Malhotra N. K., Dash S. (2011). Marketing Research an Applied Orientation. London: Pearson Publishing.
If the indicators are "sufficiently reflective" then it is probably easier to validate the reflective factor than the formative factor. So, I would keep it reflective.
Hi, James.. thanks for your invaluable videos.. one question... you mentioned the 80% rule to decide if its eventually going to be formative or reflective - what if it's a 50-50 or a 60-40? How do we decide then? Thanks in advance. .
Ideally, we decide formative/reflective measurement conceptually before we even collect data. The statistical approach is simpy a confirmation of adequate "reflectiveness" (or inter-item correlation) to justify modeling reflectively (or formatively if seeking that sort of evidence). In the case of no clear majority, then it is not sufficiently refective, and so it should either be altered to better align ti with a formative measurement approach (by removing extra dimensions - i.e., items that are not strongly correlated) or with a formative measurement approach (by removing redundant items).
Hi James, thank you very much for another highly accurate and helpful video. You provide invaluable help in the research society with your work. I would like to make two questions stemming from your video. First, you mention the 80% rule regarding the final decision whether something is formative or reflective. Is there any source for this? Second, if a reviewer asks for empirical evidence regarding a factor's measurement configuration, apart from the theoretical ones already provided, in a CB-SEM model. Is it ok to run the tetrad analysis via PLS? Is there anything similar in CB-SEM via software? I have found theoretical only references to CB-based tetrad analysis. The already very good reliability and validity properties reported via EFA and CFA are not enough for this reviewer. Thank you in advance!
1. I think I got it from this video: ua-cam.com/video/DOm_1gyiuLY/v-deo.html 2. I'm not aware of any automatic procedures for running this in CB-SEM software, although the calculation is not too difficult (if there are only four indicators). Good luck with a troublesome reviewer!
Hi James, may I ask the the source of the process for reference in validation of the models. Thank you. Usually they do bootstrapping.
All of the PLS references are on the SmartPLS website.
Hi, I planned to develop a formative model for an index, as the items of each construct are not literally or logically intercorrelated. However, after running the Confirmatory Tetrad Analysis (CTA), the results suggest that my model should be reflective, as most of the items are not significant. It seems I may need to develop a scale instead. Is passing the CTA so crucial that without it, I cannot claim this to be a formative model? Or this is only needed if I need to develop a scale, so I need to confirm this is reflective first?
The choice of formative vs reflective should be determined primarily by the design of the set of measures. If they are not intentionally correlated, then they are probably formative.
Dear James,
I ran a CTA Analysis and found that 4 of 9 cases have p-value less than 0.05, meaning 44% of all combinations and led to a formative model. Is this enough to say that the model had to be formative? I'm asking you because If I am looking at the CI low adj and CI up adj most of the correlations are -+, meaning that the model has to be reflective. Thank you very much!
Usually we lean towards a reflective model whenever reasonably possible. So, if the reliability scores and AVE are good enough, then I wouldn't even rely on the CTA. However, if you're trying to justify modeling as formative, then the CTA can provide some support. In general, it is simply easier to validate reflective models.
Hi James!!!
Is there any data set available used in this video?
Yes, data for most of the videos I make can be found on the homepage of statwiki: statwiki.gaskination.com/ This one is the Sohana dataset.
@@Gaskination thank you so much.
Thanks James!
1. You changed the arrows direction to formative type of the 3 dimensions of Burnout from Management, However, the arrows of the 6 items in the same Burnout from Management variable are still in reflective form. Could you explain the reason?
2. How is the process of determining the type of measurement model of the second-order latent variable construct? especially if we have multidimensional constructs.
1. When using the repeated indicator approach in SmartPLS for higher-order constructs that are formative, the repeated indicators should be modeled reflectively (arrows pointing at the repeated indicators). This is because the repeated indicator approach is inherently reflective in nature. (See Wetzels, M., Odekerken-Schröder, G., & Van Oppen, C. (2009). Using PLS path modeling for assessing hierarchical construct models: Guidelines and empirical illustration. MIS Quarterly, 33(1), 177-195. OR Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2017). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM). Sage Publications.)
2. The best approach is conceptual. Consider the first order dimensions and whether they are required to be correlated in a rational context. For example, is it possible for emotional exhaustion to increase while depersonalization increases? If so, then this is probably formative. If not, then this is probably reflective.
Hi James, thank you very much for your videos. after you decided unsupportive boss should be formative then the arrows were inverted for indicators not for the constructs. but for burnout management instead of arrows of indicators, arrows of constructs were inverted. could you pls explain, why?
The difference is that unsupportive boss is first order, but BM is second order.
Buenas tardes Dr. James, excelente explicación, tengo una consulta: El análisis de tétrada confirmatoria se puede aplicar para una variable de segundo orden reflexiva-reflexiva?
Creo que habría que hacerlo con las puntuaciones de las variables latentes para las dimensiones de primer orden, en lugar de con los indicadores repetidos.
Hello sir. If Ave value is coming around 0.45, even after deleting the items then what should we do? Should we report the result as it is?
Yes, report it how it is, but also check the CR value. If it is above 0.700 then this might be sufficient evidence of convergent validity. AVE is a strict measure of convergent validity. Malhotra and Dash (2011) note that "AVE is a more conservative measure than CR. On the basis of CR alone, the researcher may conclude that the convergent validity of the construct is adequate, even though more than 50% of the variance is due to error.” (Malhotra and Dash, 2011, p.702).
Malhotra N. K., Dash S. (2011). Marketing Research an Applied Orientation. London: Pearson Publishing.
@@Gaskination Thank you so much sir. God bless you sir.
Great information - thanks!
What if the CTA results show results that indicate the reflective type, whereas the theoretical framework being tested shows the formative type?
If the indicators are "sufficiently reflective" then it is probably easier to validate the reflective factor than the formative factor. So, I would keep it reflective.
Hi, James.. thanks for your invaluable videos.. one question... you mentioned the 80% rule to decide if its eventually going to be formative or reflective - what if it's a 50-50 or a 60-40? How do we decide then? Thanks in advance. .
Ideally, we decide formative/reflective measurement conceptually before we even collect data. The statistical approach is simpy a confirmation of adequate "reflectiveness" (or inter-item correlation) to justify modeling reflectively (or formatively if seeking that sort of evidence). In the case of no clear majority, then it is not sufficiently refective, and so it should either be altered to better align ti with a formative measurement approach (by removing extra dimensions - i.e., items that are not strongly correlated) or with a formative measurement approach (by removing redundant items).
Hi James, thank you very much for another highly accurate and helpful video. You provide invaluable help in the research society with your work. I would like to make two questions stemming from your video. First, you mention the 80% rule regarding the final decision whether something is formative or reflective. Is there any source for this? Second, if a reviewer asks for empirical evidence regarding a factor's measurement configuration, apart from the theoretical ones already provided, in a CB-SEM model. Is it ok to run the tetrad analysis via PLS? Is there anything similar in CB-SEM via software? I have found theoretical only references to CB-based tetrad analysis. The already very good reliability and validity properties reported via EFA and CFA are not enough for this reviewer. Thank you in advance!
1. I think I got it from this video: ua-cam.com/video/DOm_1gyiuLY/v-deo.html
2. I'm not aware of any automatic procedures for running this in CB-SEM software, although the calculation is not too difficult (if there are only four indicators).
Good luck with a troublesome reviewer!