SEM Analysis - Science method

The moderated mediation Model B refers to a statistical model where the effect of an independent variable (X) on a dependent variable (Y) is indirect, and is mediated by a mediator variable (M), but this indirect effect is moderated by one or more moderator variables (W). In the case of three moderators, the conditional indirect effect of X on Y through M can be calculated and interpreted as follows:

The results of the conditional indirect effect can be used to determine how the relationship between X, M, and Y is affected by the moderators. If the conditional indirect effect is significantly different across different levels of the moderator variables, it suggests that the moderator variables play an important role in moderating the indirect effect of X on Y through M.

Dear Naeimeh,

It is quite normal for the same MOF to have different morphologies and the morphology of the MOF is related to the synthesis conditions such as surfactant, solvent (both solvent type and solvent ratio), reactant ratio (ratio of metal to ligand), temperature, pH, reaction time etc. However, it is pretty difficult to give a general answer to this question. Crystal facet arrangement and crystal facet energy are the ultimate causes. You can try searching the relevant reviews on controlling morphology and you will get your answer.

Some related articles:

From my understanding, this shouldn't be a problem at all. With SEM you are essentially looking at covariances between variables. To this end, their absolute means as determined by the number of response options do not matter.

In general, you should not interpret the parameter estimates or fit statistics for an underidentified model. They may be completely incorrect/misleading.

SEM is a complex statistical methodology. You mention that you "know close to nothing about statistics." I would say, given that, it is not a good idea for you to use SEM unless you are guided by a statistician who is an expert in SEM and AMOS. There is a lot that can go wrong with SEM.

In your case, you have two latent variables (PerformanceExpectancy and EffortExpectancy) that each have only one indicator (observed/measured variable). These latent variables are not identified unless you add more indicators or fix the error variances of the indicators to a meaningful value. Also, your endogenous (dependent) latent variable does not seem to have an indicator (measured variable) at all.

It is impossible to know what went wrong by just looking at the input file (syntax). You need to check your data file (exp.dat) to see whether the columns/entries in that data file match with what you have in your NAMES list in the VARIABLE command. I'm almost certain that there must be some sort of mismatch between what's in the exp.dat file and the variable names list in your input file.

You can also try using a reduced set of variables (using the USEVARIABLES option) and requesting ANALYSIS: TYPE = BASIC; to check the descriptive statistics for that reduced set of variables to see if that works. That might bring you closer to identifying the problem.

Dear Ali Zia-Tohidi,

Marsh and Hocevar (1985) proposed a target coefficient T that relates the fit (Chi²) of the model with correlated first order factors to the fit of the second order factor model. Thus, the target coefficient can be used to determine whether a poor model fit is caused by the first-order models or the second-order model.

Marsh, H. W., & Hocevar, D. (1985). Application of confirmatory factor analysis to the study of self-concept: First-and higher order factor models and their invariance across groups. Psychological bulletin, 97(3), 562.

See also

Cheung, D. (2000). Evidence of a single second-order factor in student ratings of teaching effectiveness. Structural Equation Modeling, 7(3), 442-460.

best

Christoph

@ Ravisha Jayawickrama. Most sources accept value for standardised factor loadings above 0.4

Christian Geiser

Hi Christian,

Thank you very much for your quick answer.

I checked the correlations between the 3 indicators of the LS factor and they are significantly correlated according to SPSS Correlation analysis result (see picture 1).

I have estimated the same model each group separately without using multi-group analysis before and it worked well (see picture 2 and 3) (This is the reason why I say Why does an identified SEM model become unidentified after multiple group analysis with AMOS in the title of my question ). However, strangely, since I didn’t save the output at that time, I re-draw and re-run the model in two data sets (P & S groups) separately just now and find that it became unidentified and I do get the negative factor variance (-0.008, see picture 4 and 5). I'm really puzzled.

Once again thank you for your suggestion.

Best,

Bo Bo

dehydration = replacement of water by solvent (ethanol, acetone).

If you dry by evaporation of the solvent, you will obtain a very bad quality of your sample structure.

Revered Kharazmi,

Thank you for your information.

I can edit in terms of Grammatical, Editorial and Citations in your manuscript.

This is my email ID drsenapathy@gmail.com

Be in touch.

Regards

Senapathy

Ethiopia

Cesare Gabriele Gaglio

It looks like you used the following method

GelMA hydrogels were placed on copper meshes, flash frozen in liquid nitrogen and lyophilized by vacuum freeze dryer overnight. The specimen surface was then coated with gold/palladium Pt/Pd) for SEM observation.

Freeze drying is the best for this purpose. The Pt / Pd coating is necessary to protect the film from electron beams that decompose organic matter. I would try using a SEM after freeze drying, but keep the sample in the chamber for as little time as possible.

With high collinearity among predictor variables, the standardized regression coefficients may become difficult to interpret. See, for example:

XRD is a non-destructive research method (may be with the exception of biological materials), so the sample used in XRD can normally be used in other experiments (you yourself should know if the sample has been subjected of, for example, additional milling, exposure to any reagents, prolonged air storage (moisture and carbon dioxide), etc.).

Thank you, unfortunatelly I dont have any experience with this technique, but it seems plausable that you can trap them in the subtrate as they seems to be from SEM images. I just look for some papers about this technique, and it seems temperature plays a role in the agglomeration - maybe this may be your case.

To conclude, beside SEM, I would perform any chemical analysis (EDX, XPS) to be sure it is gold. Optical spectroscopy may also help, those AuNPs on quartz should have the distinguish absorbance somewhere about 600nm (blue/purple-ish look).

Hi Michael,

absolutely, blindly following modification indices can (and often will) make the model fit *better* albeit being more causally *misspecified*. Fit is only a means to an end not the end itself as completely nonsensical models can nonetheless fit very nicely.

Hayduk, L. A. (2014). Seeing perfectly fitting factor models that are causally misspecified: Understanding that close-fitting models can be worse. Educational and Psychological Measurement, 1-22. https://doi.org/10.1177/0013164414527449

Best,

Holger

Hi Rasoul,

the fact that an estimated relationship changes when stratifying (=controlling or adjusting!) a third variable, doesn't tell you anything as the key issue is whether this variable acts as a confounder (common cause auf X and Y) or collider (a common effect of X and Y). Both change the relatinship between X and Y but controlling for the confounder UN-biases the effect but controlling for the collider biases the (otherwise unbiased) effect.

I strongly recommend this book that will enlighten you :)

Pearl, J., & MacKenzie, D. (2018). The book of why. Basic books.

HTH

Holger

Unless you have a priori hypotheses about specific paths that would be expected to be non-invariant, I would think that the omnibus chi-square difference approach (comparing the fully constrained vs. unconstrained models globally) is the more appropriate approach. This is somewhat similar to what you would do in ANOVA. You look at the omnibus F test first. You run post hoc tests (pairwise comparisons) only if you obtain a significant F statistic. Otherwise you run into the risk of type-1 error inflation.

In your case with a non-significant overall chi-square difference, you would not compare individual paths.

I managed to fit my model. Working more with EFA to define cross-loadings and playing with the starting parameters to properly load the AQU and M(Social Bias) to each item.

If you thought the first model had too much going on, look at this beast!

CFI from 0.7>0.9

Joanna, Mplus now provides the option to use cat-LMS. Therefore, you just have to define the indicator variables as being categorical, and then the usual syntax of LMS can be used.

HTH, Karin

I assume that the EDS analysis was done over the area of the SEM images shown and the concentrations were obtained by pressing the “QUANT” button of the software.

The spatial resolution of EDS (laterally and in depth) is given by the beam-sample interaction (i.e. beam energy, atomic number, density) and the physics of X-ray emission.

A frequently used estimate is the Kanaya-Okayama range (K Kanaya and S Okayama, J. Phys. D: Appl. Phys. 5(1972) 43-58, DOI: 10.1088/0022-3727/5/1/308).

For pure Fe and 15 keV electrons the electron excitation range is ca. 1 µm, the size of the volume where the X-rays are from is ca. 0.7 µm for Fe K radiation.

EDS quantification assumes that the sample is sample is homogeneous, flat, and thick (compared to the excitation range).

See here: https://www.youtube.com/watch?v=et9CTYHUoos&app=desktop

This presentation by Dr. Jens Rafaelsen was recorded on the EDAX booth at M&M 2018

At 35:45 “If your sample doesn't meet those requirements: don't click the QUANT button: You'll get numbers, they'll be garbage!”

The “Fe-before” sample is likely Fe with a native iron oxide.

Is the sample flat? Is the sample homogeneous (laterally and in depth)? Is the oxide thick?

If the iron oxide is thin (< 0.7 µm), the volume you analyse is composed of iron oxide(s) + iron metal. The prerequisites of the quantification model are not fulfilled.

The “Fe-after” sample is likely Fe with thick iron oxide(s).

Is the sample flat? Is the sample homogeneous (laterally and in depth)?

The prerequisites of the quantification model are not fulfilled.

Is the oxide thick?

If the iron oxide is thick (> 1 µm), the volume you analyse might be composed of the iron oxide(s) formed, but the prerequisites of the quantification model are not fulfilled.

As mentioned by Vladimir Dusevich the EDS spectra should be evaluated carefully. There are small peaks around 2 keV. For “Fe-before” at < 2 keV, for “Fe-after” at ca. 2.3 keV. What’s this?

> Also, I have some questions about SEM-EDX:

Your questions 1-5 are not directly related to EDX.

The chemistry of oxidation of iron (and steels) in aqueous solutions has been studied for decades because of the importance for corrosion and power generation and the results are now textbook knowledge.

I guess for electrolysis a lot of actual work can be found.

For thicker oxides scales composed of different oxides may form. Here, preparation and analysis of cross sections may give more insight.

Preparation methods and analysis will depend on the thickness and stability of the scales (fracturing, preparation of cross section by metallographic methods, FIB cuts; EDS in SEM or TEM?)

I have no idea. Unfortunately this is outside of our areas of expertise. Sorry. With best wishes, Frank Edelmann

This is not necessarily meaningless. Stress is of two types: eustress and distress. Not all stress is distress and you need a certain amout of stress= eustress to feel well

Ali A. Al-Allaq: I believe Mehraj Din Wanie 's question is about structural equation modeling (a statistical technique), not microscopy.

Although there are some rough guidelines and rules of thumb in the literature as to the minimum sample size required for SEM, those guidelines lack generality because the sample size depends on many different factors (e.g., model size, number of indicators, factor loadings/reliability of indicators, effect sizes, amount of missing data, normality of data, etc.). In my opinion, the best way to determine the optimal sample size is by running a Monte Carlo simulation with your specific model and the expected parameter values. That way, you can increase the certainty that the sample size estimate will actually apply to your specific case. For some guidance, see

I offer a free mini-course on sample size planning in SEM that you can find here:

These indices are fairly outdated anyway. It is more common nowadays to look at the chi-square test of model fit, RMSEA, CFI or TLI, and SRMR. Guidelines were provided by Hu and Bentler (1999) as well as Schermelleh-Engel, Moosbrugger, and Mueller (2003). My personal opinion is that the most relevant information about model (mis)fit is provided by the chi-square test and the model residuals.

Hello Rowena,

I would not trust any arbitrary threshold values of some magical indices. Better is to statistically test your model with a chi-square test. And, yes, there is also a arbitrary threshold involved but this threshold (the .05 p-value) does not tell your when the model is problematic but represents a cultural agreement when to regard a deviation from chance as noticable.

If you tell more bout the model (including exact question wordings and intended meaning of the latents) and tell the model chisquare test and df, I could possibly help :)

All the best

Holger

P.S. One note: Of course you can invent abbreviations like CLF as you like but this makes conversations difficult.

First - NFC are too small to be accurately observed with SEM. You need to use AFM for those fibers. The coating depth required is not insignificant and will obscure any details that you may have with larger diameter NFC fibers. As cellulose fibers are hygroscopic, it is very unlikely that you will have a completely dry sample for SEM. The technique that I've found works is to allow the sample to dry in an oven at 50C, then store it in a nitrogen atmosphere until ready to coat. Once coated you should have no issues obtaining good scans of your fibers. I coat approximately 4 nm depth and obtain good resolution images.

Marco Marini as far as I know, you must have two conditions both verified for a partial mediation hypothesis to be confirmed:

1 - the indirect effect must be significant (X -> M -> Y) *

2 - the direct effect must be significant (X -> Y)

If both conditions are satisfied, then you have a partial mediation. If condition 1 is satisfied, but not condition 2, then you have a full mediation (i.e., your mediator entirely explains the effect of X over Y).

As Christian Geiser suggested: "Partial mediation simply means that only some of the X --> Y effect is mediated through M".

To my knowledge, the ratio between direct and indirect effect has no role in distinguishing between partial vs. full mediation.

* Please note: "the indirect effect must be significant" doesn't mean that path a and b must both be significant. All you need is path a × b significant (better if bootstrapped).

Model fit assessment should be based on a lot more information than just the indices that you mentioned. First of all, I would pay close attention to the chi-square test of model fit and it's p value. If the test is significant, looking at model residual statistics is often useful to determine sources of misfit. Descriptive fit statistics such as RMSEA and CFI tend to be less sensitive to model misspecification, especially in larger samples. Also, a low CFI may indicate that the variables in your model have rather low correlations on average as this index compares the target model fit to the fit of a baseline independence model.

Also important to look at are the parameter estimates and their standard errors. Are the parameter estimates all proper with the expected sign and do the standard errors look reasonable in terms of their magnitude?

Observation of Microstructure and Mechanical Properties in

Heat Affected Zone of As-Welded Carbon Steel by Using

Plasma MIG Welding Process

It was seen that the use of plasma MIG welding process has resulted in the refinement of the microstructure in the CGHAZ region, thus improving the mechanical properties of the as-welded SPCC steel. The highest microhardness values were obtained for conventional MIG in the CGHAZ regions. Incorporation of the plasma arc reduces the hardness, potentially increasing the ductility of the joining by plasma MIG welds. Further reduction in hardness can be obtained by decreasing the plasma current values.

In general, carbon adhesive tape is not good for high-resolution SEM, because it usually induces some drift. The easiest way to prepare a stable SEM samples for bulk materials (powders, granules, etc) - just put some conductive adhesive (just a thin layer) on the surface of a clean stub, and then immediately pour the powder. When the adhesive dries, just remove the excess of the material with a stream of pure air or nitrogen. That's it.

Good luck!

Thankyou fir the answer. I'll try this method

Hello Seyyed,

I guess you mean with beta the factor loading? Traditionally, these are denoted with lambda but probably, Stata treats these differently.

The fixation of the "marker variable" is needed a) to assign a metric to the latent variable--those of the marker, and to b) identify the equation system.

As far as I know it does not matter which variable you choose unless it is no valid indicator of the latent.

HTH

Holger

If you have a cells monolayer, 30 min is a good time. If you have something like a tissue developing, with a lot of collagen, then you need 1 hr. HA is soluble in water (very slow, but still...). So if you culture started generate small centers of mineralization, you do not want to keep it too long (days, weeks) in water solutions. From the other side prolonged storage in desiccator can lead to fungus growth. Some desiccators are badly infested with fungus and need through cleaning and disinfection. From my opinion the best way to store specimens is when their preparation is complete, i.e. they are dehydrated and coated with conductive coating.

I have also come across this situation. In one of my papers, I used EFA and then the structural model. Recently received review comments that I should run CFA. The reviewers recommended that if you have a valid scale to measure then you should run CFA must. EFA can only run under the exploration stage only. As many experts have already answered this question, let me ask a little more about this:

1. When reliability and validity can be checked by Cronbach alpha and expert validity, still is it mandatory to run CFA to check validity and reliability?

2. If exploratory research extracts factors with their items and loadings, why do we further need to run CFA?

Looking forward to hearing from experts.

Hi there,

a completely different approach could be to run a TETRAD model. This allows to set some restrictions where you can be sure about the direction (e.g., the autoregressive effects as well as forbidding reverse effects from t_n to t_n-1) and freely expore the rest. The model will print a path diagram that shows you three things

1) clearly supported causal (happens only rarely)

2) Effects with one side (the arrowhead) crealy "depentent" but the other end ambiguous (may be a cause or a consequent)

3) completely ambiguous relationships

TETRAD exists since the 80s and is remarkably invisible to our field.

Eberhardt, F. (2009). Introduction to the epistemology of causation. Philosophy Compass, 4(6), 913-925.

Malinsky, D., & Danks, D. (2018). Causal discovery algorithms: A practical guide. Philosophy Compass, 13(1), 1-11. https://doi.org/10.1111/phc3.12470

A few final comments

1) 90% of confirmatory SEMs are much too complex and involve dozens and sometimes hundreds of testable implications. And because of that the models never fit which--in almost funny manner--is then used as support for the model ("well, models never fit, so why should my do?"). I would always focus one ONE or TWO essential effects or chains of effect and then try to a) think hard about confounders and b) potential instruments

2) Yes, causation needs time to evolve but most often, the time lag is embedded in the measurement. Otherwise you would not have any cross-sectional correlations. That is, if the causal lag is similar to the lag embedded in the measure (e.g., "how satisfied are you with your job" will prompt an answer that is derived from memory) OR / AND the IV is stable, then cross-sectional data will generally allow to identify causal effects. The key issue is and stays "causal identification"--that is removing confounding biases and potential reverse effects. The latter can be solved in a cross-lagged design but not the former. That is, you have to think hard about confounding no matter what the timely design is.

I had a long discussion in the following thread in case you differ in your opinion (which is fine for me):

Best,

Holger

"Effect size - f2 " tells whether a construct has a substantive impact on another one. Guidelines for assessing ƒ2 are (Cohen, 1988): values of 0.02, 0.15, and 0.35, respectively, represent small, medium, and large effects of an exogenous latent variable on an endogenous latent variable.

Thank you Francisco Arteaga for the fascinating question and professors for the valuable answers, I agree with your answers.

Social researchers have access to numerous statistical methods. Consequently, knowing the proper technique can be difficult. For instance, selecting between covariance-based (CB-SEM) and variance-based partial least squares (PLS-SEM) can be difficult when considering structural equation modelling (SEM). It conducts a direct comparison using the same theoretical measurement and structural models and data set. In order to achieve an acceptable goodness-of-fit when employing CB-SEM as opposed to PLS-SEM, a greater number of indicators are eliminated. In addition, composite reliability and convergent validity were typically higher when PLS-SEM was utilized, whereas other metrics such as discriminant validity and beta coefficients were comparable. PLS-SEM performed significantly better than CB-SEM when comparing variance explained in indicators of the dependent variable. The revised guidelines aid researchers in determining whether CB-SEM or PLS-SEM is the most appropriate technique to employ.

Reference

Hair Jr, J. F., Matthews, L. M., Matthews, R. L., & Sarstedt, M. (2017). PLS-SEM or CB-SEM: updated guidelines on which method to use. International Journal of Multivariate Data Analysis, 1(2), 107-123.

Samples were fixed by adding 5% glutaraldehyde solution overnight which was replaced with a fresh sterilised solution of PBS and changed three times before soaking in fresh sterilised deionised water for one hour twice. Then, the CCC samples were frozen at −80 °C for three hours before placing them into a Christ ALPHA 2–4 freeze-dryer for 24 hours.

Srikanth Satish Kumar Darapu Be aware that the mineralogy is likely to change with particle size. Softer, more brittle minerals are found in the finer fractions and the harder (gangue) materials may be found in the larger size fractions.

Hello Hyunsoon,

Shifting from estimated factor scores as an IV to individual constituent manifest variable scores as IVs will do several things (none of which is particularly good):

1. Your comparison (does a path exist among latent variables in sem vs. does some linear combination of manifest variables relate to some other score, regardless of whether it is for a latent or manifest variable) is no longer "apples to apples" or the same research question, so no inference may be made as to whether one is better than another.

2. Regressing your "B" score on multiple A manifest indicators almost guarantees that the weights which might apply from your sem measurement model will be ignored in order to maximize the multiple R/R-squared in your regression. Hence, a relationship may or may not help make the case for the constructs being related.

3. The approach is somewhat like modifying the model in order to yield the results you would like to see, and so any interpretation of p-values (for a model changed after having looked at a prior analysis of the same data) is likely misleading, and the opportunity for overfitting is higher.

Finally, please note that model fit indices are driven by how well estimated model parameters serve to reproduce the observed correlations/covariances among the measured variables. You certainly can have good fit with no significant paths among latent variables, if the latent variables are unrelated.

Good luck with your work.

Hello Edita,

A categorical variable having only two levels (e.g., coded 0/1) can be used in any linear model as an IV or antecedent variable.

If such a variable is the DV, however, it likely makes more sense to switch from linear to logistic models.

Good luck with your work.

The models are nested. You can see this from the fact that an equivalent way to specify Model 2 by setting two factor correlations in Model 1 to 1.0, namely between Factors 1 and 2 as well as 4 and 5.

There are two complications in this case, however, when it comes to applying nested model chi-square testing. First, correlation coefficients of 1.0 are at the boundary of the admissible parameter space for correlations, violating assumptions of the chi-square test. Stoel et. al. (Psychological Methods) developed a correction for this. The second complication is the non-normality correction of the chi-square values for each model. You would have to take the scaling correction factor into account when computing chi-square difference tests. It is perhaps easier to simply compare the absolute fit of the models or look at information criteria such as BIC.

Dear Thathsarani,

there are five types of variables in any causal system (besides your target independent variable X and outcome Y):

1) Confounders or variables lying on a "backdour path". Confounders are variables that act as common causes of X and Y and create a spurious/non-causal connection between X and Y when you not adjust/control for them. This non-causal connection is in particular covariance that is not due to an X-->Y effect and can create and effect where there is non at all or bias the effect. The bias can be an upward bias or downwad bias depending on the X-effect and the confounding effect (it is quite easy to predict, google for "path tracing").

A backdoor path is the path from X on Y that passes the confounder (see above). If a variable is part of this path but innocent in the confounding process (where the confounder is the one guilty), adjusting for such a variable will, too, eliminate the bias. That is, even if you cannot control for the confounder itself, controlling fo the "surrogate confounder" will do the job.

THIS what a control variable has to achieve.

2) Mediators are variables that transmit an effect of X on Y. Never accidently control for them unless you delibaretely want to eliminate the indirect effect (e.g., to exclude a certain process and you want to see if there are other processes or a residual direct effect). That is: take a look onto your set of predictors/control variables, and check whether there could be a mediator among them that you missed in the first place. Controling for a mediator causes overcontrol bias (see Elwert for a nice example)

3) Moderators (better term: effect modifyer) are variables which affect the effect of X on Y or in other words: The moderators present some kind of circumstance or context in which the effect may occur (or not) or that affect the size or direction of that effect. For instance, take a simple example of the X "banging with your head to the wall" which results in the Y "headache". However, this effect is moderated by the dummy variable "helmet" (yes/no).

A moderator CAN also be a confounder (affecting X and Y) OR a mediator (transmitting some portion of the X-Y effect). The key issue is whether it affects the residual X-Y effect that results from simply controlling for the moderator.

4) Instrumental variables, finally, are variables which either a) are a cause X but have no direct effect on Y or b) have a relationship with X (but not Y) that is due to a confounder. Instrumental variables are valuable as they can achieve to identify a causal X-Y effect (in a two-stage-least squares regression or a SEM, see Maydeu-Olivares et al., 2019) even if there there is confounding and you cannot control for the confounder or variables on the backdoor path.

Simply (mis)using instruments as controls will make a confounding bias worse! This is call Z bias or bias amplification effect.

5) Colliders are variables which are a common EFFECT of X on Y (the opposite of confounders). Controlling for a collider accidently will lead to a bias. The surprising thing: Collider bias also happens when you control for the Y (or select some subgroup, or stratify Y) or an outcome of Y! To illustrate that, take a look onto this model (by the way: A brilliant browser software!) and move the mouse over Y (not click) and type "A" (for adjust) on your keyboard.

This simulates an adjustment process. The bias is signalled by the effect on X becoming purple. Now repeat this (after de-typing A) to Z and see the same result.

The solution of this miracle is the error terms of Y and Z. To get an unbiased effect, there must not be a correlation between X and the e's. Y, however is a collider in the X-e1 path, and Z is a collider in the X-e1 path. Hence, controlling for Y and Z creates the bias. In graph theoretical parlance, controlling a collider "opens a closed path"--creating a spurious relationship.

I know this is a lot to swallow but really understanding these variables helps a lot.

It should be noted that to evaluate on the role of these type of variables is a purely theoretical issue. The data will not tell you that.

If you have any question, please ask.

All the best,

Holger

Maydeu-Olivares, A., Shi, D., & Rosseel, Y. (2019). Instrumental variables two-stage least squares (2SLS) vs. maximum likelihood structural equation modeling of causal effects in linear regression models. Structural Equation Modeling: A Multidisciplinary Journal, 26(6), 876-892.

Elwert, F. (2013). Graphical causal models. In S. L. Morgan (Ed.), Handbook of causal analysis for social research. (pp. 245-273). Springer. https://doi.org/10.1007/978-94-007-6094-3

Your CFA/SEM fit statistics and standard errors may be biased due to non-normality when using standard maximum likelihood estimation. In case of non-normality, you can use maximum likelihood estimation with robust test statistics and standard errors such as the Satorra-Bentler correction to avoid/reduce bias.

You can try for modified Poulton’s reagent to etch this alloy. Use freshly prepared reagent for better results. The composition of modified Poulton’s reagent is as follows:-

50ml Poulton reagent + 25ml HNO3 +1ml HF+1ml H2O.

(Poulton reagent : 12mlHCl concentrated+6mlHNO3+1ml HF(48%)+1mlH2O).

time 5 to 10 s

Agreed, multigroup CFA/SEM is a powerful method for group comparisons with regard to both measurement-related issues and structural (latent variable) parameters. Regarding your second question, I like to use Mplus for multigroup CFA and SEM as it now conveniently automates measurement invariance (MI) testing across groups. When using the setting

ANALYSIS: MODEL = CONFIGURAL METRIC SCALAR;

Mplus will automatically estimate those three models with different levels of MI and compare their fit via chi-square difference tests. I have a free Youtube video here in which I demonstrate this: https://www.youtube.com/watch?v=0vVv-TiJp-4&t=27s

Dear all, indeed complexing agents have additional roles on the morphology, shape, size and size distribution. Various studies are devoted to witness on their benificial role. Please have a look at the following documents. My Regards

doi: 10.1021/la903470f

DOI:10.3390/batteries6040048

Lucky Sibanda I think that sounds useful - having a model that includes factors in existing literature to compare their relative effects. In terms of pre-enrolment factors, I think Tinto's suggestion of controlling for students' commitment to persist with education from pre-enrolment is important. We have developed a model for measuring this which hopefully we will put a working paper out on soon. We identified subfactors like social pressure to persist, expectation mismatch, and a couple of other things. Let me know if you are interested and maybe I can share it with you. Some aspects of commitment to persist will be culturally specific like in some regions and class groups very high proportions of people go to higher ed so there is more of a social pressure to get a degree even for those who do not really want to. I think accounting for structural factors pre and post enrolment is very important before looking at micro-sociological or psychological factors. Things like family or personal hardship are probably going to be more predictive than any psychological factor but maybe a model that accounts for structural factors can identify what psychological factors or social supports are protective after a factor that is out of an institution's control has been accounted for.

Dear Yinan Li

Please see Table 1 and relevant studies/stages in the following recently published JR article. The article also provides an extensive web appendix, which should help you with EFA and CFA in a PLS-SEM based study. Please note that this is an open/free access article. Good luck!

Syed Mahmudur Rahman, Jamie Carlson, Siegfried P. Gudergan, Martin Wetzels, Dhruv Grewal. (2022). Perceived Omnichannel Customer Experience (OCX): Concept, measurement, and impact. Journal of Retailing, https://doi.org/10.1016/j.jretai.2022.03.003

Hi Omar,

Please see Table 1 and relevant studies/stages in the following recently published JR article. The article also provides an extensive web appendix, which should help you with PLS-SEM based analysis and reporting. Please note that this is an open/free access article. Good luck!

Syed Mahmudur Rahman, Jamie Carlson, Siegfried P. Gudergan, Martin Wetzels, Dhruv Grewal. (2022). Perceived Omnichannel Customer Experience (OCX): Concept, measurement, and impact. Journal of Retailing, https://doi.org/10.1016/j.jretai.2022.03.003

Jochen Wilhelm I agree very much with your statement that "you should better do more research on the meaning of the variable you are actually analyzing." I think that this is generally desirable for many studies. I also agree that there is a tendency in the social sciences to overemphasize standardized coefficients and to not even report unstandardized coefficients. That is very unfortunate in my opinion, as I believe both are important and have their place.

That being said, there are fields (mine included: psychology) where we are dealing with variables that simply do not have an intuitive metric. Many variables are based on test or questionnaire sum or average scores. People use different tests/questionnaires with different metrics/scoring rules in different studies. What does it mean when, for example, subjective well-being is expected to change by 2.57 for every one-unit change in self-rated health and by 1.24 for every one-unit change in extraversion when self rated health is measured on a 0 - 10 scale and extraversion ranges between 20 and 50?

Standardized estimates can give us a better sense for the "strength" of influence/association in the presence of other predictors than unstandardized coefficients when variables have such arbitrary and not widely agreed upon metrics. The interpretation in standard deviation (SD) units is not completely useless in my opinion, especially since we operate a lot with SD units also in the context of other effect size measures such as Cohen's d. It allows us (often, not always) to see fairly quickly which variables are relatively more important as predictors of an outcome--we may not care so much about the absolute/concrete interpretation or magnitude of a standardized coefficient, but it does matter whether it is .1 or .6.

In addition, in the context of latent variable regression or path models (i.e., structural equation models), unstandardized paths between latent variables often have an even more arbitrary interpretation as there are different ways to identify/fix latent variable scales (e.g., by using a reference indicator or by standardizing the latent variable to a variance of 1). Regardless of the scaling of the latent variables, the standardized coefficients will generally be the same.

This does not mean that I recommend standardized coefficients over unstandardized coefficients. Variance dependence and (non-)comparability across studies/different populations are important issues/drawbacks of standardized coefficients. Unstandardized coefficients should always be reported as well, and they are very useful when variables have clear/intuitive/known metrics such as, for example, income in dollar, age, number of siblings (or pretty much any count), IQ scores, etc. Unstandardized coefficients are also preferable for making comparisons across groups/populations/studies that used the same variables. I would always report both unstandardized and standardized coefficients along with standard errors and, if possible, confidence intervals.

I believe there are many examples of regression or path models in psychology for which standardized coefficients were reported and that did advance our knowledge regarding which variables are more important than others in predicting an outcome.

GUI type -https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwja6ezvsfr2AhX1zTgGHRL0C1AQFnoECD4QAQ&url=https%3A%2F%2Flink.springer.com%2Fcontent%2Fpdf%2F10.1007%252Fs00180-012-0317-1.pdf&usg=AOvVaw31Y9KGVr-UESpSL097bzkL

Thank you so much again, Jessica Tay Ying Ling

It really helped.

I am no expert in plant microscopy, but you still do not have relevant answers... In general you can store biological specimens for a while in a fixative solution (glut) in a refrigerator. Or fix specimens, dehydrate properly in graded solutions of alcohol and stop at 70-96% of alcohol and store in a refrigerator.

@Imran Thank you

Yes, you can use the same sample for more techniques, but those techniques should not modify the sample, meaning that those techniques should not be invasive. In your case, you can use the same sample for UV-Vis, XRD, SEM, and FTIR.

try to move to PLS or SPSS regression analysis

I am disappointed that I am late for the party - a very interesting discussion for someone trying to understand SEM like myself. Waqar Ashiq I guess you are done with your study (it's impressive that you had to read widely and eventually improved your model).

Your question of negative loadings has been answered from the discussion above, I suppose. After looking at your two models, I have some questions, perhaps some people who are part of the discussion will clarify here ( Zainudin Awang , Christian Geiser , Marcel Grieger , Karin Schermelleh-Engel , David Eugene Booth )

I will refer to your initial model as Model 1 and the revised model as Model 2.

1. On Model 1, why did you include an error term on the Topographical latent independent variable (exogenous)? I think all the other latent variables should have an error term except this one.

2. Just like the above point raised, on Model 2, there should be no error term on the Landscape latent exogenous variable. Waqar Ashiq , I will be glad to get a source that justifies an error term on an independent latent variable.

3. On Model 1, I see many error terms of observed variables covariated while they form part of different latent variables. For example, e1 is covariated with e9 while they form part of two different latent variables Topographical and Biological respectively. I agree with what Imtiaz Ahmad mentioned. However, covariating e1 and e2 is appropriate since they are both under the latent variable Topographical.

4. Is it advisable to use 2 items per latent variable in a SEM model (like in both models above)? I read somewhere that 3 or more are better.

5. In Model 1, independent latent variables are not covariated. Is it optional to covariate independent latent variables?

I listened to a course on YouTube by Saeed Sharif: https://www.youtube.com/watch?v=-rGIRIPgeWY&list=PLdQI483_ZTKzhaaIGjswUxkHNAbX1Qi78&index=5&ab_channel=saeedsharif

Your model is overidentified with 6 degrees of freedom. This means that there are six testable restrictions in the model: six possible direct paths between variables are currently not estimated (these paths are implicitly fixed at zero). The high chi-square value indicates that some of these "not estimated"/fixed-to-zero direct paths/effects may be non-zero in the population. Are there any possible direct paths that you omitted in your model that may be non-zero? You can examine this by looking at standardized covariance residuals (these residuals can show you which observed covariances/correlations are underestimated by the model) and/or by looking at model modification indices. The modification indices may point you to those additional paths that should be freely estimated since they differ significantly from zero.

The model cannot properly be estimated via regression because it is overidentified (non-saturated, has > 0 degrees of freedom).

I think the bottom line of the reviewer comment is that the cross-sectional design is inadequate for examining the focal research question or hypotheses on causal effects. This is not primarily a stats issue but a matter of research methods in more general terms.

I have seen lots of submissions to academic journals applying sequential designs. That is, researchers introduce a time lag of a few weeks between measuring predictor and criterion, but do not include repeated measures of the focal variable (no panel data). In my view, such designs are a more or less desparate attempt to avoid desk rejection because of cross-sectional data.

There is quite a bit research that suggests that sequential designs do not perform better than cross-sectional data (see below).

I agree that experimental or real longitidunal survey data are a much better fit to test causal effects. In my view, providing evidence for causality is possible in limited ways to impossible with non-experimental data. I can recommend the chapter of John Antonakis discussing among other things the role of third variables. I like his example of a loud noise occurring (instead of a rifle) causes discs to be shattered, because the correlation between the occurrence of a crack and the condition of the discs is so highly correlated. :-)

In my view, leaving the causality obsession (in psychology) aside, there are lots of meaningful and important research questions that can be addressed with correlational data (even cross-sectional data), e.g. regarding the interplay of variables or the relative weight of predictors (no matter if these predictors are the only or real causes).

Best

Oliver

Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. (2014). Causality and Endogeneity. The Oxford Handbook of Leadership and Organizations. https://doi.org/10.1093/oxfordhb/9780199755615.013.007

O’Laughlin, K. D., Martin, M. J., & Ferrer, E. (2018). Cross-Sectional Analysis of Longitudinal Mediation Processes. Multivariate Behavioral Research, 53(3), 375–402. https://doi.org/10.1080/00273171.2018.1454822

Mitchell, M. A., & Maxwell, S. E. (2013). A Comparison of the Cross-Sectional and Sequential Designs when Assessing Longitudinal Mediation. Multivariate Behavioral Research, 48(3), 301–339. https://doi.org/10.1080/00273171.2013.784696

I thinking, when there is no significant difference found Through, its improper to use it as control variable. Better is allow PLS to create its own groups if any were present in the data.

XRD analysis defines the crystal structure that is the formation of desired phase (s). The peak analysis represents the volume of unit cell. In case of nanomaterial, crystallite size can also be computed. In contrast, the SEM analysis represents surface structure, the grain, grain boundary, and porosity.

In briefly, XRD provides internal information of a material while SEM gives sole the surface information.

If ion mill (as described by Pierre Caulet ) is out of reach, you can use metallographic polishing. To preserve thin layers it's better to make specimen "sandwich": with thin epoxy glue two pieces of specimen with substrate on outside. Embed, polish.