Here's a fun pet project I've been working on: udreamed.com/. It is a dream analytics app. Here is the RUclips channel where we post a new video almost three times per week: ruclips.net/channel/UCiujxblFduQz8V4xHjMzyzQ Also available on iOS: apps.apple.com/us/app/udreamed/id1054428074 And Android: play.google.com/store/apps/details?id=com.unconsciouscognitioninc.unconsciouscognition&hl=en Check it out! Thanks!
Hi James, Is it acceptable for the significance effect of 0.07? Can I say that there is a weak relationship between 2 variables because it is close to 0.05? I really don’t want to reject my initial hypotheses. Thanks again
The chosen confidence level is up to you. Many choose 90% confidence. It is only customary to choose 95%, but there is no rule about it. Instead, ensure you have adequate statistical power and then check the effect size of the relationship. If it is meaningful, then you can argue that your hypothesis is supported, even if it is only at the 90% confidence level.
@themanager1980 The general rule of thumb is to have 10 times the number of rows as columns. So if you have 5 variables, you need a sample size of 50. This is a generous rule though. You can get as low as half that and still be okay. So, if you have 5 columns, you need at least 25 responses on each variable. Personally, I feel that a sample size of less than about 60 is at high risk of type II errors (not being able to detect significant relationships that actually do exist).
Hi Sir, thank you very much for your video. I have a question: usually, to run a model in Amos, it is required to draw the constructs connected with observed variables (items used to measured that construct), otherwise i coundnt run it because Amos always shows a notification of erros relating to observed variable (when i only use the elip) or unobserved variable (when i only used the rectangle). Even when using your plugin Pattern matrix model builder it also shows the constructs with observed variables. But here in your video, i can see that you only use the rectangle to indicate the construct (instead of the elip) without any observed variables, and still it works. Could you please explain why and guide me how to do that! Thank you very much!
In this video, I'm using factor scores resulting from a CFA. So, each of these rectangles represents a weighted and standardized average of the latent factor ellipse.
How do you report regression analysis results (like the one in this video) conducted with Amos in APA style? Do you report only standardized regression weights, p values and total R2? As I get it R2 in the output is Squared Multiple Correlations, numbers like B, S.E., Beta, t, ΔR etc can be found somewhere in amos output? Do you have to check model fit as well?
wonderful... I only have 5 columns and 8 rows I applied same as your way...But are the results significant? I cannot get more than 5 variables and 8 samples. Thanks
Thanks very much, Dr. Gaskin. I still have some questions: 1) Is the C.R. the one given in the table which has "Regression Weights"? Or is there a C.R. value for the model as a whole? 2) What do you suggest should be reported in the results of a simple linear regression done in AMOS? Should it be the B, beta, S.E., significance of B, R-square and C.R.? Or have I missed something?
Hello James, I am trying to run a CFA on Amos but when i click the calculate estimates button, the model doesnt run. All i get is Scanning (Filename)...Reading data....Please what might be the problem?
If you mean it never finishes, then it could be a slow computer or a very complex model or too many options checked in the output tab of the analysis properties, or a combination of all of these.
alamindabo1 Hi Alamindabo1, I am also encountering the same problem. Were you able to solve it? I would greatly appreciate any help. thank you very much.
I'm not sure I understand the question. If you want to export one of the output tables to a spreadsheet, just left click next to it, then right click and copy. There is no "save" or "export" function for the output.
Thank you very much for all of your videos, they are very helpful, easy to follow and to understand, the best in the net. Also thanks very much for the reply. Which video shows the model fit for this kind of regression (without latent variables) beacause I have seen the model fit for a model with latent variables but I can't find something similar to this one?
You cannot covary endogenous variables (variables with arrows pointing into them). AMOS won't let you. If absolutely necessary, you can covary their error terms (implying that the error from the regression line is systematically correlated - possibly due to similarity of wording in the survey questions). Hope this helps.
Sir, I have a doubt while I go through it. It's showing that "an error occured while attempting to fit the model. A sample size was not specified. "So that I can't calculate estimate.
Missing data can be imputed if less than 20% is missing. You can impute with the median or using other more sophisticated techniques. SPSS now has a Missing Values analysis feature. I should make a video on it... If a specific row or variable is missing more than 20% of the values, it becomes much less valid to impute the values.
James the issue is I want use an article and there was no effect size reported and fir some reason I should rely on the results if this article what they reported was only unstandarized beta and I have to get effect size from this so is there any way ?
Dear Gaskin, In spss one have to check F statistic. Does it not matter to check fit index when using amos to do multiple regression? In fact when I did this, I found chi square and df being a zero but this does not any sense to me unfortunately. In fact i resort to Amos for having 2 dependent variable which are higly correlated, so I would like to control each other. Can I report coefficient without giving fit index because these are 0 but I also very high RMSEA. My feelings tell me I am on the wrong way.
The f-statistic is not reported in AMOS. The chi-square and DF are zero when all possible paths are accounted for. When all paths are accounted for, you have "perfect fit" because the observed model perfectly matches the predicted model.
Hello James, I followed your septs reading CFA issue with negative error variances "This shouldn’t happen if all data screening and EFA worked out well, but it still happens… In such cases, it is permitted to constrain the error variance to a small positive number (e.g., 0.001)" the loading of this indicator show to be equal to 1.00 (standard estimation), is it normal?
+Riad Cheikh Standardized loadings should be between -1 and 1, so it is possible for it to be 1. However, you might consider constraining the indicator paths to be equal (I'm guessing there are only two or three indicators). You can do this by naming the paths the same thing in the parameter constraint area (double-click the path). Then add a variance constraint to the latent factor. This should fix it.
+James Gaskin Thx James, Would doing what you have said on the error variance (naming the error variances to be the same name instead of constrained 0.001) that would fix the problem of 1.00, is it statistical correct?
+James Gaskin I meant constrain the Variance of error variance to be 0.001 makes the loading of this indicator to be 1.00, I tried to use the way you described in how to solve the Heywood cases by naming the constrain variance to the same name (AAA) for all error variances for the same latent variable
Thank you so much for your reply. I am afraid my question was not clear. I would like to do the following: Test a model with say 4 latent constructs, and compare it with a alternative model with say 8 constructs,using other indicators, different paths, testing different theory (eg. non nested) I am using a single group. I understand I can use AIC or ECVI (to compare groups) But the question is. I must do that in separe files or can do it using different models in AMOS ? Thank you so much
Depends. If you have some ideas about how the variables should be grouped (by concept) then go ahead and do some theoretical groupings, then do some factor analysis to confirm it. Or, if this is entirely exploratory (sort of like cluster analysis), then jump right into factor analysis and see what comes out. Then return to the wording of the variables so that you can figure out what is the underlying common theme within a factor.
Missing values will be accounted for differently. SPSS uses a different "replacement" method for handling missing data. AMOS uses an estimation method via Means and Intercepts.
It is tricky, but you can do it by clicking on the status bar on the bottom left of the window that says "not estimating any user-defined estimand" and then selecting to create a new estimand. You'll have to google how to do the rest though, and it will help if you are familiar with programming logic.
Dear James, I am again stuck on another front of AMOS SEM. The Dof is positive still my model is unidentified. It is asked to add 2 additional constrains. How to resolve this?Notes for Model (Default model) Computation of degrees of freedom (Default model) Number of distinct sample moments: 78 Number of distinct parameters to be estimated: 29 Degrees of freedom (78 - 29): 49 Result (Default model) The model is probably unidentified. In order to achieve identifiability, it will probably be necessary to impose 2 additional constraints. SE11
Usually this occurs when you have regression constraints missing on latent factors. So, each latent factor should have one indicator path with a regression constraint to 1. Make sure there is one path per factor that is constrained. (double click the path, go to parameters, set regression weight to 1.
Hi Mr James. thanks a lot for SEM presentations. I am wondering if you can help me, I am trying to do a path analysis in AMOS. every time i tried to run the software a note pups up saying that the endogenous variables have no residual (error) variables. I do not know how overcome this problem. Many Thanks
The endogenous variables (the ones being predicted, your Dependent Variables), need to have error terms. The error term button is the one that looks like a rectangle with a circle over it. Click on it and then click on the endogenous variable. I show how to do this at time point 1:15 in this video.
Thanks James. My DV represents 2 groups where a value above or below determines to which group a case belongs. Would it be better to just use multiple regression with the DV being the value change? If so, how many IVs can I use to maintain sufficient power?
I would use logistic regression. Logistic regression is specifically designed to predict group membership. Your statistical power will be determined largely by the sample size of each group.
I am trying to validate a survey using CFA. I have one data set that I will be using for EFA and CFA. Do I depend on EFA to assign variables to a factor on CFA or is there any other way? Thanks.
If the measures have previously been validated in extant literature, then you can use those validated sets to inform how you build your CFA. If instead, these are new measures that you have created, then you should run the EFA and let that help inform your CFA.
@@anar2636 Here are several methodological papers on CFA: statwiki.gaskination.com/index.php?title=References#Confirmatory_Factor_Analysis If instead you are simply looking for an implementation of a CFA in a standard research paper, here is one of mine: decisionsciences.org/wp-content/uploads/2017/11/p746959.pdf
think you have to enter variables in blocks i.e. hierarchical multiple regression. in such situation, it gives the relationship only between the dependent variable and independent variables. then how do you check the relationship between the variables in different blocks? i.e variables in 1st block and 2nd block, 2nd block and the 3rd etc. have to use several regressions ?
Hi James, thanks for your great video. Can you tell me why you covary the independent variables (role conflict, work uncertainty, and dead-end job in the structural model by adding paths between them? Just to improve model fit or any reason else? Can I find any reference for this procedure? Thanks.
Can I use SEM with 3 independent variables (identified through PCA), 1 binary dependent variable, and N=60? I am using data on a neighborhood level (similar to census tract), which is why my N is somewhat small.
Yes, you can do this, but since it is a binary DV, you have to interpret the output a little differently. The regression weights between the IVs and DV are read more like a t-test or ANOVA, where a positive regression weight means that the higher value of the binary DV is associated with higher values of the IV, and a negative regression weight means that the lower value of the binary DV is associated with higher values of the IV. The better approach is probably just to create factor scores (or composite values) for the latent factors during the EFA or CFA, and then conduct a logistic regression in SPSS.
Again your videos are the best. Could you please give an example of the above explanations? I am not really sure I understand. I have innovation as binary DV, and network as binary IV. They have negative regression. how do i interpret it ?
If you mean how to impute missing values, you do that in SPSS using the "replace missing values" function in the "transform" menu. If you mean how to impute composite variables from latent variables, that is done in AMOS using the "Data Imputation" function in the Analyze menu.
Reverse mediation? Do you mean inverse? Like where the indirect effect is negative? If so, it is handled the same way. The way to interpret it is just to recognize that the effect of the IV on the DV through the mediator is a negative effect.
This is a assumption for latent models in covariance based sem. As far as I understand it, the algorithm requires one constrained path per latent factor. It won't run without that constraint (unless the constraint is placed on the variance of the latent factor instead).
Hi, I have a large discrepancy between the r2 originated by SPSS amos (r2=0.73) and by the multiple regression analysis (r2=0.33), Do you have any explanation? The discrepancy appeared only for one variable, the others are completely fine. Many Thanks
Sir.........In factor analysis there is the option to save score as variable in spss for factors. I just only want to know....is there any option or way for the same in CFA by using AMOS
Once the SEM is complete, what function exports it to Word. Copy function does not seem to work. In my SEM model, Theta score is the latent variable for a population of students and I have 6 independent valuables which based on my hypothesis will or will not predict a diiference in the theta score. Do I have add the student to the model as an linear or horizontal connection in the model?
I cannot understand your first question. Do you want to know where the residual values are in AMOS? If so, they are in the S.E. column of the tables in the output. The value on top of the dependent variables is the R2.
Hi Prof Gaskin, Thank so much from this video. I'm quite new in SEM and AMOS. Please can tell me how to recognize the latent variables among my variables. Should I run a factor analysis first in order to specify the latent variables or I should just pick them from the theory. I look forward to hearing back from you. Many thanks
Hey James, i just did an analysis with AMOS (SEM). The Problem: i got a correlation of over 1 (-1,60) between the IV and a Mediator. What could be a problem ? Greetings Roman
This rarely happens if the data is clean and normally distributed, and if the EFA and CFA worked. If your data isn't suited to EFA/CFA, then make sure it is also not categorical. If categorical, then it needs to be broken up into binary dummy variables.
@@Gaskination Thank you very much :) It was non-standardized :D my fault. Is there any possibilty to do a SEM ? There is some heteroskedasticity in my data. I used for the linear regression a robust estimator. Is there something like this in AMOS or any other programm ? Greets
@@romanpauck3453 AMOS doesn't do robust estimators, but Mplus does. I don't have a video for MLR though. Essentially, you just need to add the ESTIMATOR statement and then type MLR;
@@romanpauck3453 In Mplus, you have many statements, such as DATA: or VARIABLE: So, for the ANALYSIS statement, include ESTIMATOR = MLR; Check out page 11 here: users.ugent.be/~yrosseel/lavaan/utrecht2010.pdf
standardized is what we use because they are comparable across contexts and across different variable sizes. The effects will be universally understandable since they are always between -1 and +1.
Very important and useful page Prof. Gaskin ! One question. I would like to compare alternative models in AMOS, in the same way I can draw different models for different groups (checking the "Allow different path diagrams for different groups"). is there any way to do the same with different models ? I can only compare different models in regard to constraints , but not to path or variables. Is that possible ?
Hi James thanks for an amazing job this is the first tie mi use Amos to run regression and im trying to confirm my SPSS out put findings i followed this video and i applied the same procedures exactly but i got horrible TFI, CFI and RMSEA could you please tell me why and how to fix it ???
James you man Awesome! thanks i have another question can i convert the unstandarized coeffiecient beta into r, d, or even OR ?! any thing that help in calculating the sample size thanks in advance
James (e.g. Gaskination): Can you please share how to correlate dependents? For example, let's say we had 5 dependents such as OCB. We can't draw arrows to order them but can't we correlate them? How can we do this?
The reason for the difference in this video is due to the different set of variables used. I used dead end job in the Amos model but I used resource demand gap in the SPSS model. Regressions are sensitive to the set of variables included in the model. If I used the same set of variables, I would find very similar results between the two analyses. The point is, adding or dropping variables will change the effects of other variables.
Thank you very much, James. But which analysis should one choose in such a situation if they just duplicate each other: AMOS model, linear regression or just correlations between dependent and independent variables? Or by another way, which additional information gives each method comparative to each other?
@@smaximok1 If there is only one dependent variable, then linear regression or amos model should be identical (or very close). Correlations will be different because they are bivariate and do not account for common variance across multiple predictors.
Hello Dr Gaskin, Thank you for your videos. Those are very helpful. I would like to know if I can run path analysis with moderation in AMOS. Also, do I have to run CFA or just straight away do regression?
+Aishath Nasheeda You can do path analysis with moderation. I have many videos on this. If you have reflective latent factors, then you must first run EFA and CFA before doing path analysis.
Hi James, I am using SPSS.... And i face some problem here... I wish to conduct regression analysis on two variable, let it as A and B Both A and B in my questionnaire each contains few others questions to increase it reliability For instance A consist of Q1 and Q2 While B consists of Q3, Q4, and Q5 So when i key in the results into the SPSS, i key in as in Q1, Q2, Q3, Q4, and Q5 And now my problem is, how do i use SPSS to do regression analysis between A (Q1 and Q2) and B(Q3, Q4, Q5) as the dependent variable only allowed me to insert one, for instance Q1 only... Hope to hear from you soon... Thanks
I use AMOS instead. If you look at my SEM Series playlist on my channel, you'll see how to do such an analysis. Or there is also a video that's called "from measurement to structural model" ruclips.net/video/n-ULF6BGVw0/видео.html which shows how you might do this. If you don't have AMOS, then you'll either have to create averages for those items, or you'll have to create factor scores with a factor analysis in spss (see my EFA video).
hi james, i want to analze a likert scale questionerre having 8 factors each having 4 questions, how can i do the correlation analysis between 2 factors. I could not do it from SPSS. Because it allows only one variable for correlation in my situation it becomes a question not a factor.
It depends on the complexity of your model. Ideally you will have a minimum of five subjects per variable in your model. It can be done with less, but then you sacrifice statistical power (the ability to detect significant relationships that do exist). Hope this helps. (p.s. it's "Structural" not "Standard")
Oh. You can run a homogeneity of variance test (Levene's Test) in SPSS. Just create a new variable that identifies which year each row is from, then do the homogeneity of variance test using that new variable as the factoring variable.
I am so thankful James for your help. Another thing why we do confirmatory factor analysis ? if we have variables which are mostly single items. in that case can we use structural equation modelling?
James Gaskin what does mean this error amos will require the following pairs of variables to be uncorrelated , I try to run it but it show this message and below variable 1 variable 2
Hani Albarni It means you have some exogenous variables that are not covaried. You can proceed, but it is expected you will covary all exogenous variables.
James Gaskin Ok fine I was able to create the model now my question where do I can find correlation output ? what do I need to select from estimates option ?!
Hello doctor. Thanks for great videos. I really appreciate your efforts. Can u help me in analysis. If I want to see the change in sales before and after the covid19. I have collected the daily sales data from same retailer. Then what test I can use. Can you please suggest please.
You mean you want to download AMOS? IBM sells it. You can get a discounted version on the academicsuperstore website if you are a student or faculty. I don't distribute it.
systematic means rule-based or methodical. in our case, it refers to the correlation (relationship between two variables) varying together for the same reason (perhaps because of similarity of survey question structure or wording). Hope this helps.
Yes. Once you are done drawing the model, you can always click on the white stack icon to drag a variable over a box. This will insert that variable in the box.
Doctor, Your help please , i have a problem , some constructs are having AVE less than 0.5, i thought it is bcz i insert zero for the " not aaplicable- NA" in likert scale 🥺 Your opinion please 🙏🙏🙏🙏🙏
Yes, zero will break it. Instead, delete the zero values and rerun. This will require estimating means and intercepts (due to missing data) if you're using AMOS.
Hello Prof Gaskin, I am testing a model in which one of the exogenous variables is dichotomous (Yes=1 and No=2), other are on likert scale, can I use this dichotomous as regressor?
Yes, but it makes more sense to have Yes be the higher number and No be the lower number. That will make interpretation simpler. In this case, I would recommend changing all your 2 to zero so that No is zero and Yes is 1.
Not sure what PLA is, unless you mean PLS. Lisrel and AMOS are covariance based SEM modeling software. PLS is an algorithm separate from covariance based algorithms for SEM. SmartPLS, pls-graph, and WarpPLS use the PLS algorithm. Typically we model reflective constructs in AMOS or Lisrel, but we do formative ones in PLS software. Many arguments have also been made to do exploratory models in PLS.
Thank you so much for this video (and the others too, obviously!) Is there a way in AMOS to group multiple variables so as to act as a single one? Sorry if I lack the technical term, but English is not my language of study, and I do not know the corresponding term.
Hi, Thanks for the video. It's very concise and useful. Does AMOS have an equivalent of the 'F' value of SPSS? What should we include in the results of a simple linear regression done in AMOS?
Hey! Could you help me to understand how to download the model of results? I mean that I could make a screenshot but maybe there is one more way to download it accurately...thanks!!
The export to Word and Excel might work (they don't work for me). You can find these in the output window at the top. If those don't work, then you can highlight the section you'd like to export, then click on the copy icon (next to the open icon) on the output page. Then paste this in Word.
Hi James, can AMOS run multiple regression? And is it true that one of the criteria of using AMOS is that it's compulsory to have 10-likert scale? Thanks
Many thanks for your video. I learned a lot about Amos just in 5 min. About the statistics, I am wondering why you mention we apply means and intercept if we have missing values?! Means and intercept coming directly from research question and model specification. is not so?
The standardized loadings should be between +/- 1.00. Unstandardized loadings have no bounds. A large standardized loading is above 0.500, moderate above 0.300, and small above 0.150.
Oh. To copy the model, just use a screen capture tool. If you are using windows 7 or 8, just type "Snip" into the search area and you'll find the snipping tool.
Dear James, Many thanks for your fabulous videos, I have a question which I really appreciate your advice. when performing the chi-square test at the end to compare my measurement models with structural model (SEM and CFA), my measurement model and structural model have the exact same DF and Chi-squares therefore the difference is zero with chi-square of zero, so is it acceptable?
This will sometimes happen when you have no degrees of freedom. A chi-square test only works when you have degrees of freedom. If you do have degrees of freedom, and it is still identical, then it is simply because you have regressed everything upon everything, which is essentially the same (for model fit) as covarying everything (like in the CFA).
James Gaskin Thanks james, you are Fantastic I do have DF, so I guess then I need to perform other tests right? or is it not acceptable to have regressed everything upon everything? ... appreciate your advice. On another note, what would you do if your chi-square test is significant? do I need to keep running CFA until the chi-square test works? what remedies you suggest? p.s. my supervisor is totally against covarying error terms so I need to omit variables due to MI or look at factor loading basically to improve my model.
Salman heydari khajehpour It is very uncommon to have a non-significant p-value for the chi-square, especially if you have a complex model or a large sample size. I almost never report that measure. I usually stick with RMSEA, CFI, cmin/df. I can't remember if your variables are all ordinal, but if not, if they are continuous, then normality affects model fit as well.
AMOS is CB SEM, so it assumes you will covary all exogenous variables, but it is not a requirement to get it to run. The only thing you really sacrifice when you don't covary them is you are unable to get correlation values between the constructs you do not covary.
One other thought on this. If you have multiple DVs, then they are being estimated simultaneously with the other DVs. SPSS doesn't do this, so in AMOS, some of the variance being explained in one of the DVs may actually cannibalize the explained variance in another. In SPSS they are run independently.
Hi, Dr. Gaskin, could you please tell me why the result is different between the analysis done by AMOS and SPSS? I mean in your amos demo, wc was not significant, but in your spss demo, resource gap something was not significant while wc was significant. Actually, i usually build a structural model based on multiple regression analyses. but in some model, i found that the regression paths that I hypothesized are not significant even though they were significant in regression analyses with SPSS.
Hello James! Your videos are really helpful. I have a question, maybe you could help. I’m gonna use spss for my thesis analysis. Could I use the regression presented in this video to analyze the influence of independent variables (UTAUT Model) on behavioral intention to adopt a technology? Thank you
.sav is from SPSS. If you don't have that, .csv should still work. You just need to go to File, Data Files, and then click on File Name, and then change the file type to csv (the dropdown just above the open and cancel buttons).
James, thanks for the great video. I have a simple question. Do I check the regression weights table for the size of the effect or the standardized regression weights table ? because the estimates are different in these tables
The unstandardized values are the actual slope or coefficient for the regression equation. The standardized values always fall between -1 and 1 and so they are easy to compare across multiple paths and multiple studies. If you are in the social sciences, use the standardized effects. If you are in the hard sciences, use the unstandardized effects. I'm in Information Systems (a social science), so I use the standardized effects.
James Gaskin YOu are great ! just great ! so i will use the standardized regression weights table which are also same with the numbers that i get on the model. James last question please... Did you run path analysis or simple regression in this video ? best regards,
Hi. thanks for the helpful video. I am very new to statistics so I'm sorry if this is a stupid question... would this regression you have done in the video count as Structural Equation Modelling?
Here's a fun pet project I've been working on: udreamed.com/. It is a dream analytics app. Here is the RUclips channel where we post a new video almost three times per week: ruclips.net/channel/UCiujxblFduQz8V4xHjMzyzQ
Also available on iOS: apps.apple.com/us/app/udreamed/id1054428074
And Android: play.google.com/store/apps/details?id=com.unconsciouscognitioninc.unconsciouscognition&hl=en
Check it out! Thanks!
I am a PhD student, I found it really Very useful, it was really very useful. Your language speed every thing was perfect
Thank you
Understood. I am specifically copying the model not the output. Your information has been helpful
Hi James, Is it acceptable for the significance effect of 0.07? Can I say that there is a weak relationship between 2 variables because it is close to 0.05? I really don’t want to reject my initial hypotheses.
Thanks again
The chosen confidence level is up to you. Many choose 90% confidence. It is only customary to choose 95%, but there is no rule about it. Instead, ensure you have adequate statistical power and then check the effect size of the relationship. If it is meaningful, then you can argue that your hypothesis is supported, even if it is only at the 90% confidence level.
James Gaskin thanks again James.
@themanager1980
The general rule of thumb is to have 10 times the number of rows as columns. So if you have 5 variables, you need a sample size of 50. This is a generous rule though. You can get as low as half that and still be okay. So, if you have 5 columns, you need at least 25 responses on each variable. Personally, I feel that a sample size of less than about 60 is at high risk of type II errors (not being able to detect significant relationships that actually do exist).
Thank you for making this look so simple and easy to understand. You are doing a great job!!!!
Thank you so much, our professor sucked at explaining this but you make it effortless!
Hi Sir, thank you very much for your video. I have a question: usually, to run a model in Amos, it is required to draw the constructs connected with observed variables (items used to measured that construct), otherwise i coundnt run it because Amos always shows a notification of erros relating to observed variable (when i only use the elip) or unobserved variable (when i only used the rectangle). Even when using your plugin Pattern matrix model builder it also shows the constructs with observed variables. But here in your video, i can see that you only use the rectangle to indicate the construct (instead of the elip) without any observed variables, and still it works. Could you please explain why and guide me how to do that! Thank you very much!
In this video, I'm using factor scores resulting from a CFA. So, each of these rectangles represents a weighted and standardized average of the latent factor ellipse.
Very straightforward and easy to understand.
How do you report regression analysis results (like the one in this video) conducted with Amos in APA style? Do you report only standardized regression weights, p values and total R2? As I get it R2 in the output is Squared Multiple Correlations, numbers like B, S.E., Beta, t, ΔR etc can be found somewhere in amos output? Do you have to check model fit as well?
wonderful... I only have 5 columns and 8 rows I applied same as your way...But are the results significant? I cannot get more than 5 variables and 8 samples.
Thanks
It is the same for structural models as for measurement models. The same measures, the same procedures, and the same thresholds.
Thanks very much, Dr. Gaskin. I still have some questions:
1) Is the C.R. the one given in the table which has "Regression Weights"? Or is there a C.R. value for the model as a whole?
2) What do you suggest should be reported in the results of a simple linear regression done in AMOS? Should it be the B, beta, S.E., significance of B, R-square and C.R.? Or have I missed something?
This is great, I always wonder why nobody made statistical videos, thanks you soooooo MUUUUUCH
Hello James,
I am trying to run a CFA on Amos but when i click the calculate estimates button, the model doesnt run. All i get is Scanning (Filename)...Reading data....Please what might be the problem?
If you mean it never finishes, then it could be a slow computer or a very complex model or too many options checked in the output tab of the analysis properties, or a combination of all of these.
alamindabo1
Hi Alamindabo1, I am also encountering the same problem. Were you able to solve it? I would greatly appreciate any help. thank you very much.
I'm not sure I understand the question. If you want to export one of the output tables to a spreadsheet, just left click next to it, then right click and copy. There is no "save" or "export" function for the output.
Thank you very much for all of your videos, they are very helpful, easy to follow and to understand, the best in the net. Also thanks very much for the reply. Which video shows the model fit for this kind of regression (without latent variables) beacause I have seen the model fit for a model with latent variables but I can't find something similar to this one?
Is there any way to find different results due to different estimation of missings that these two programs use by default?
You cannot covary endogenous variables (variables with arrows pointing into them). AMOS won't let you. If absolutely necessary, you can covary their error terms (implying that the error from the regression line is systematically correlated - possibly due to similarity of wording in the survey questions). Hope this helps.
Sir,
I have a doubt while I go through it.
It's showing that "an error occured while attempting to fit the model. A sample size was not specified. "So that I can't calculate estimate.
I can only imagine this is due to some error with the data. Check your linked dataset to make sure there are no issues with it.
those things do go hand in hand. it is not sure proof, but systematic correlation is a necessary (but not sufficient) condition of both.
Hi, what do you suggest has to be done initially for missing data before we perform the analysis?
Missing data can be imputed if less than 20% is missing. You can impute with the median or using other more sophisticated techniques. SPSS now has a Missing Values analysis feature. I should make a video on it... If a specific row or variable is missing more than 20% of the values, it becomes much less valid to impute the values.
James the issue is I want use an article and there was no effect size reported and fir some reason I should rely on the results if this article what they reported was only unstandarized beta and I have to get effect size from this so is there any way ?
Dear Gaskin, In spss one have to check F statistic. Does it not matter to check fit index when using amos to do multiple regression? In fact when I did this, I found chi square and df being a zero but this does not any sense to me unfortunately. In fact i resort to Amos for having 2 dependent variable which are higly correlated, so I would like to control each other. Can I report coefficient without giving fit index because these are 0 but I also very high RMSEA. My feelings tell me I am on the wrong way.
The f-statistic is not reported in AMOS. The chi-square and DF are zero when all possible paths are accounted for. When all paths are accounted for, you have "perfect fit" because the observed model perfectly matches the predicted model.
@@Gaskination thanks so much
Hello James,
I followed your septs reading CFA issue with negative error variances "This shouldn’t happen if all data screening and EFA worked out well, but it still happens… In such cases, it is permitted to constrain the error variance to a small positive number (e.g., 0.001)"
the loading of this indicator show to be equal to 1.00 (standard estimation), is it normal?
+Riad Cheikh Standardized loadings should be between -1 and 1, so it is possible for it to be 1. However, you might consider constraining the indicator paths to be equal (I'm guessing there are only two or three indicators). You can do this by naming the paths the same thing in the parameter constraint area (double-click the path). Then add a variance constraint to the latent factor. This should fix it.
+James Gaskin Thx James, Would doing what you have said on the error variance (naming the error variances to be the same name instead of constrained 0.001) that would fix the problem of 1.00, is it statistical correct?
+Riad Cheikh Not naming the error variances, but naming the indicator paths.
+James Gaskin I meant constrain the Variance of error variance to be 0.001 makes the loading of this indicator to be 1.00, I tried to use the way you described in how to solve the Heywood cases by naming the constrain variance to the same name (AAA) for all error variances for the same latent variable
Thank you so much for your reply. I am afraid my question was not clear.
I would like to do the following:
Test a model with say 4 latent constructs, and compare it with a alternative model with say 8 constructs,using other indicators, different paths, testing different theory (eg. non nested)
I am using a single group. I understand I can use AIC or ECVI (to compare groups) But the question is. I must do that in separe files or can do it using different models in AMOS ? Thank you so much
Depends. If you have some ideas about how the variables should be grouped (by concept) then go ahead and do some theoretical groupings, then do some factor analysis to confirm it. Or, if this is entirely exploratory (sort of like cluster analysis), then jump right into factor analysis and see what comes out. Then return to the wording of the variables so that you can figure out what is the underlying common theme within a factor.
Missing values will be accounted for differently. SPSS uses a different "replacement" method for handling missing data. AMOS uses an estimation method via Means and Intercepts.
It is tricky, but you can do it by clicking on the status bar on the bottom left of the window that says "not estimating any user-defined estimand" and then selecting to create a new estimand. You'll have to google how to do the rest though, and it will help if you are familiar with programming logic.
Dear James, I am again stuck on another front of AMOS SEM. The Dof is positive still my model is unidentified. It is asked to add 2 additional constrains. How to resolve this?Notes for Model (Default model)
Computation of degrees of freedom (Default model)
Number of distinct sample moments: 78
Number of distinct parameters to be estimated: 29
Degrees of freedom (78 - 29): 49 Result (Default model)
The model is probably unidentified. In order to achieve identifiability, it will probably be necessary to impose 2 additional constraints.
SE11
Usually this occurs when you have regression constraints missing on latent factors. So, each latent factor should have one indicator path with a regression constraint to 1. Make sure there is one path per factor that is constrained. (double click the path, go to parameters, set regression weight to 1.
Hi Mr James. thanks a lot for SEM presentations. I am wondering if you can help me, I am trying to do a path analysis in AMOS. every time i tried to run the software a note pups up saying that the endogenous variables have no residual (error) variables. I do not know how overcome this problem. Many Thanks
The endogenous variables (the ones being predicted, your Dependent Variables), need to have error terms. The error term button is the one that looks like a rectangle with a circle over it. Click on it and then click on the endogenous variable. I show how to do this at time point 1:15 in this video.
Thanks a lot. really appreciated
Thanks James. My DV represents 2 groups where a value above or below determines to which group a case belongs. Would it be better to just use multiple regression with the DV being the value change? If so, how many IVs can I use to maintain sufficient power?
I would use logistic regression. Logistic regression is specifically designed to predict group membership. Your statistical power will be determined largely by the sample size of each group.
Great. Thank you!
I am trying to validate a survey using CFA. I have one data set that I will be using for EFA and CFA. Do I depend on EFA to assign variables to a factor on CFA or is there any other way? Thanks.
If the measures have previously been validated in extant literature, then you can use those validated sets to inform how you build your CFA. If instead, these are new measures that you have created, then you should run the EFA and let that help inform your CFA.
@@Gaskination Thanks. Could you share with me a paper where in you have interpreted the CFA results.
@@anar2636 Here are several methodological papers on CFA: statwiki.gaskination.com/index.php?title=References#Confirmatory_Factor_Analysis
If instead you are simply looking for an implementation of a CFA in a standard research paper, here is one of mine:
decisionsciences.org/wp-content/uploads/2017/11/p746959.pdf
think you have to enter variables in blocks i.e. hierarchical multiple regression. in such situation, it gives the relationship only between the dependent variable and independent variables. then how do you check the relationship between the variables in different blocks? i.e variables in 1st block and 2nd block, 2nd block and the 3rd etc. have to use several regressions ?
I don't use blocks very often (only when doing control variables in SPSS), so I'm not sure. Sorry! Best of luck to you.
Hi James, thanks for your great video. Can you tell me why you covary the independent variables (role conflict, work uncertainty, and dead-end job in the structural model by adding paths between them? Just to improve model fit or any reason else? Can I find any reference for this procedure? Thanks.
Can I use SEM with 3 independent variables (identified through PCA), 1 binary dependent variable, and N=60? I am using data on a neighborhood level (similar to census tract), which is why my N is somewhat small.
Yes, you can do this, but since it is a binary DV, you have to interpret the output a little differently. The regression weights between the IVs and DV are read more like a t-test or ANOVA, where a positive regression weight means that the higher value of the binary DV is associated with higher values of the IV, and a negative regression weight means that the lower value of the binary DV is associated with higher values of the IV. The better approach is probably just to create factor scores (or composite values) for the latent factors during the EFA or CFA, and then conduct a logistic regression in SPSS.
Oops. I was logged in under one of my other channels... TheAISChannel is me...
Again your videos are the best. Could you please give an example of the above explanations? I am not really sure I understand. I have innovation as binary DV, and network as binary IV. They have negative regression. how do i interpret it ?
If you mean how to impute missing values, you do that in SPSS using the "replace missing values" function in the "transform" menu. If you mean how to impute composite variables from latent variables, that is done in AMOS using the "Data Imputation" function in the Analyze menu.
Reverse mediation? Do you mean inverse? Like where the indirect effect is negative? If so, it is handled the same way. The way to interpret it is just to recognize that the effect of the IV on the DV through the mediator is a negative effect.
Thank you for this demonstration, this is very useful. I do have a question--where are df for the SEM?
degrees of freedom show up in the 'notes for the model' section of the output in AMOS.
hello sir,
why we have regression weight 1 for one path in the each measurement model set ,
it comes by default so does it make any sense.
This is a assumption for latent models in covariance based sem. As far as I understand it, the algorithm requires one constrained path per latent factor. It won't run without that constraint (unless the constraint is placed on the variance of the latent factor instead).
thanks
Hi, I have a large discrepancy between the r2 originated by SPSS amos (r2=0.73) and by the multiple regression analysis (r2=0.33), Do you have any explanation? The discrepancy appeared only for one variable, the others are completely fine.
Many Thanks
Sir.........In factor analysis there is the option to save score as variable in spss for factors.
I just only want to know....is there any option or way for the same in CFA by using AMOS
Once the SEM is complete, what function exports it to Word. Copy function does not seem to work.
In my SEM model, Theta score is the latent variable for a population of students and I have 6 independent valuables which based on my hypothesis will or will not predict a diiference in the theta score. Do I have add the student to the model as an linear or horizontal connection in the model?
I cannot understand your first question. Do you want to know where the residual values are in AMOS? If so, they are in the S.E. column of the tables in the output. The value on top of the dependent variables is the R2.
Hi Prof Gaskin,
Thank so much from this video.
I'm quite new in SEM and AMOS. Please can tell me how to recognize the latent variables among my variables. Should I run a factor analysis first in order to specify the latent variables or I should just pick them from the theory.
I look forward to hearing back from you.
Many thanks
Hey James,
i just did an analysis with AMOS (SEM). The Problem: i got a correlation of over 1 (-1,60) between the IV and a Mediator.
What could be a problem ?
Greetings
Roman
This rarely happens if the data is clean and normally distributed, and if the EFA and CFA worked. If your data isn't suited to EFA/CFA, then make sure it is also not categorical. If categorical, then it needs to be broken up into binary dummy variables.
@@Gaskination
Thank you very much :)
It was non-standardized :D my fault.
Is there any possibilty to do a SEM ? There is some heteroskedasticity in my data. I used for the linear regression a robust estimator. Is there something like this in AMOS or any other programm ?
Greets
@@romanpauck3453 AMOS doesn't do robust estimators, but Mplus does. I don't have a video for MLR though. Essentially, you just need to add the ESTIMATOR statement and then type MLR;
@@Gaskination
Hey, thanks for the reply ! :)
What do you exactly mean be "Estimator statement" ? I'm not a native english speaker :/
@@romanpauck3453 In Mplus, you have many statements, such as DATA: or VARIABLE: So, for the ANALYSIS statement, include ESTIMATOR = MLR; Check out page 11 here: users.ugent.be/~yrosseel/lavaan/utrecht2010.pdf
Fast and to the point..lovely
standardized is what we use because they are comparable across contexts and across different variable sizes. The effects will be universally understandable since they are always between -1 and +1.
Very important and useful page Prof. Gaskin !
One question. I would like to compare alternative models in AMOS, in the same way I can draw different models for different groups (checking the "Allow different path diagrams for different groups"). is there any way to do the same with different models ? I can only compare different models in regard to constraints , but not to path or variables. Is that possible ?
Hi James thanks for an amazing job this is the first tie mi use Amos to run regression and im trying to confirm my SPSS out put findings i followed this video and i applied the same procedures exactly but i got horrible TFI, CFI and RMSEA could you please tell me why and how to fix it ???
Great videos, thank you. Will it be possible to share the data that you have used in this and the other videos?
+Piyush I share many datasets on the homepage of my wiki: statwiki.kolobkreations.com
James you man Awesome! thanks i have another question can i convert the unstandarized coeffiecient beta into r, d, or even OR ?! any thing that help in calculating the sample size thanks in advance
James (e.g. Gaskination): Can you please share how to correlate dependents? For example, let's say we had 5 dependents such as OCB. We can't draw arrows to order them but can't we correlate them? How can we do this?
Hello James, can I use dummy variables in AMOS as an independent variables? Thank you. Best
The reason for the difference in this video is due to the different set of variables used. I used dead end job in the Amos model but I used resource demand gap in the SPSS model. Regressions are sensitive to the set of variables included in the model. If I used the same set of variables, I would find very similar results between the two analyses. The point is, adding or dropping variables will change the effects of other variables.
Thank you very much, James.
But which analysis should one choose in such a situation if they just duplicate each other: AMOS model, linear regression or just correlations between dependent and independent variables? Or by another way, which additional information gives each method comparative to each other?
@@smaximok1 If there is only one dependent variable, then linear regression or amos model should be identical (or very close). Correlations will be different because they are bivariate and do not account for common variance across multiple predictors.
Hello Dr Gaskin, Thank you for your videos. Those are very helpful. I would like to know if I can run path analysis with moderation in AMOS. Also, do I have to run CFA or just straight away do regression?
+Aishath Nasheeda You can do path analysis with moderation. I have many videos on this. If you have reflective latent factors, then you must first run EFA and CFA before doing path analysis.
Hi James,
I am using SPSS.... And i face some problem here...
I wish to conduct regression analysis on two variable, let it as A and B
Both A and B in my questionnaire each contains few others questions to increase it reliability
For instance A consist of Q1 and Q2
While B consists of Q3, Q4, and Q5
So when i key in the results into the SPSS, i key in as in Q1, Q2, Q3, Q4, and Q5
And now my problem is, how do i use SPSS to do regression analysis between A (Q1 and Q2) and B(Q3, Q4, Q5)
as the dependent variable only allowed me to insert one, for instance Q1 only...
Hope to hear from you soon...
Thanks
I use AMOS instead. If you look at my SEM Series playlist on my channel, you'll see how to do such an analysis. Or there is also a video that's called "from measurement to structural model" ruclips.net/video/n-ULF6BGVw0/видео.html which shows how you might do this.
If you don't have AMOS, then you'll either have to create averages for those items, or you'll have to create factor scores with a factor analysis in spss (see my EFA video).
Thanks james... u just made my day....
Hi James. How can i change the format of file, to .sav? because my data mostly in word and excel
You would have to import it into Spss and then save it as an Spss file. Spss uses the default file type .sav.
@@Gaskination ohh...thank you for reply💫
Dear Mr. James, Is it possible to analyze SEM using Partial Least Square?
Thanks
How do you create a graphical output of your derivations ? Great video
You can take a screenshot or create it in powerpoint
The C.R. (Critical Ratio) is something like the F value. Needs to be greater than 1.96 in order to be significant.
Hey James, do we always need to covariate the independent variables?
In covariance-based software, like AMOS, the independent variables should always be covaried.
And also has a moderating effect to test the relationship between two groups. Which software? Plz
If nesting of sub dependent and sub sub dependent variable is there. then what should we do?
Im in the middle of my thesis this video help alot! thanks!!
hi james, i want to analze a likert scale questionerre having 8 factors each having 4 questions, how can i do the correlation analysis between 2 factors. I could not do it from SPSS. Because it allows only one variable for correlation in my situation it becomes a question not a factor.
classicguitarr Watch my videos on Exploratory factor analysis, and also on confirmatory factor analysis.
It depends on the complexity of your model. Ideally you will have a minimum of five subjects per variable in your model. It can be done with less, but then you sacrifice statistical power (the ability to detect significant relationships that do exist). Hope this helps.
(p.s. it's "Structural" not "Standard")
Oh. You can run a homogeneity of variance test (Levene's Test) in SPSS. Just create a new variable that identifies which year each row is from, then do the homogeneity of variance test using that new variable as the factoring variable.
I am so thankful James for your help. Another thing why we do confirmatory factor analysis ? if we have variables which are mostly single items. in that case can we use structural equation modelling?
CFA is only for latent (multi-item) factors. So, if all single item, then no CFA.
Hello Mr.James good morning, where do i can find regression output in AMOS after going to choose from Output ?
It is in the Estimates --> Scalars section. It is labeled as Regression Weights Table.
James Gaskin what does mean this error amos will require the following pairs of variables to be uncorrelated , I try to run it but it show this message and below variable 1 variable 2
Hani Albarni It means you have some exogenous variables that are not covaried. You can proceed, but it is expected you will covary all exogenous variables.
James Gaskin Ok fine I was able to create the model now my question where do I can find correlation output ? what do I need to select from estimates option ?!
Hani Albarni Make sure to check the box for standardized estimates in the output tab of the analysis properties window.
Hello doctor. Thanks for great videos. I really appreciate your efforts. Can u help me in analysis. If I want to see the change in sales before and after the covid19. I have collected the daily sales data from same retailer. Then what test I can use. Can you please suggest please.
Dear James, would you pls. give me a link to downloal the extension of SEM at AMOS ??
You mean you want to download AMOS? IBM sells it. You can get a discounted version on the academicsuperstore website if you are a student or faculty. I don't distribute it.
systematic means rule-based or methodical. in our case, it refers to the correlation (relationship between two variables) varying together for the same reason (perhaps because of similarity of survey question structure or wording). Hope this helps.
But can i first design the model then insert the data variable that i have drawn?
Yes. Once you are done drawing the model, you can always click on the white stack icon to drag a variable over a box. This will insert that variable in the box.
I am unable to click on my output path diagram buttin. It has been disabled but I can view my text (view text)
Doctor,
Your help please , i have a problem , some constructs are having AVE less than 0.5, i thought it is bcz i insert zero for the " not aaplicable- NA" in likert scale 🥺
Your opinion please 🙏🙏🙏🙏🙏
Yes, zero will break it. Instead, delete the zero values and rerun. This will require estimating means and intercepts (due to missing data) if you're using AMOS.
You saved my life doctor, thanks alot 🙏🙏🙏🙏🙏🙏
I did not delete them, instead i make the 0 as a missing in spss which is the same i think 🙏🙏🙏
@@ahmed3820 Yes, it is the same unless you have zeros in your dataset that are legitimate values.
I don't understand I think. Why would you want to calculate the sample size from the beta? You can't get a beta without having some sample first...
Hello Prof Gaskin, I am testing a model in which one of the exogenous variables is dichotomous (Yes=1 and No=2), other are on likert scale, can I use this dichotomous as regressor?
Yes, but it makes more sense to have Yes be the higher number and No be the lower number. That will make interpretation simpler. In this case, I would recommend changing all your 2 to zero so that No is zero and Yes is 1.
Thanks Prof Gaskin, you are a life savor!!
Not sure what PLA is, unless you mean PLS. Lisrel and AMOS are covariance based SEM modeling software. PLS is an algorithm separate from covariance based algorithms for SEM. SmartPLS, pls-graph, and WarpPLS use the PLS algorithm. Typically we model reflective constructs in AMOS or Lisrel, but we do formative ones in PLS software. Many arguments have also been made to do exploratory models in PLS.
how we can control for some variable like size etc in structural equation modelling
Just add it as another independent variable. Add a line from it pointing to whatever variable you think it may affect (usually the DV).
thanks your comments are so much helpful
Thank you so much for this video (and the others too, obviously!) Is there a way in AMOS to group multiple variables so as to act as a single one? Sorry if I lack the technical term, but English is not my language of study, and I do not know the corresponding term.
Hi,
Thanks for the video. It's very concise and useful.
Does AMOS have an equivalent of the 'F' value of SPSS? What should we include in the results of a simple linear regression done in AMOS?
Good, thanks very much. This is a nice clear introductory presentation of AMOS.
Hey! Could you help me to understand how to download the model of results? I mean that I could make a screenshot but maybe there is one more way to download it accurately...thanks!!
The export to Word and Excel might work (they don't work for me). You can find these in the output window at the top. If those don't work, then you can highlight the section you'd like to export, then click on the copy icon (next to the open icon) on the output page. Then paste this in Word.
Hi James
In the tools menu, what can I do with the (Seed Manager and Golden) ?
Thanks in advance
Alshimaa Ibrahem Don't worry about those. They refer to the way the randomization algorithms work.
James Gaskin Thank you
Described in a really simple manner - great
Hi James, can AMOS run multiple regression? And is it true that one of the criteria of using AMOS is that it's compulsory to have 10-likert scale? Thanks
The path model shown in this video is a form of multiple regression. AMOS can handle any size scale, or even continuous data.
Thanks! Appreciate your reply :)
Many thanks for your video. I learned a lot about Amos just in 5 min. About the statistics, I am wondering why you mention we apply means and intercept if we have missing values?! Means and intercept coming directly from research question and model specification. is not so?
This is just a constraint of AMOS. AMOS cannot run with missing data unless this box is checked.
May I know is there any specific figure for path loading. One of mine was very high.
The standardized loadings should be between +/- 1.00. Unstandardized loadings have no bounds. A large standardized loading is above 0.500, moderate above 0.300, and small above 0.150.
Oh. To copy the model, just use a screen capture tool. If you are using windows 7 or 8, just type "Snip" into the search area and you'll find the snipping tool.
Dear James,
Many thanks for your fabulous videos, I have a question which I really appreciate your advice. when performing the chi-square test at the end to compare my measurement models with structural model (SEM and CFA), my measurement model and structural model have the exact same DF and Chi-squares therefore the difference is zero with chi-square of zero, so is it acceptable?
This will sometimes happen when you have no degrees of freedom. A chi-square test only works when you have degrees of freedom. If you do have degrees of freedom, and it is still identical, then it is simply because you have regressed everything upon everything, which is essentially the same (for model fit) as covarying everything (like in the CFA).
James Gaskin
Thanks james, you are Fantastic
I do have DF, so I guess then I need to perform other tests right? or is it not acceptable to have regressed everything upon everything? ... appreciate your advice.
On another note, what would you do if your chi-square test is significant? do I need to keep running CFA until the chi-square test works? what remedies you suggest?
p.s. my supervisor is totally against covarying error terms so I need to omit variables due to MI or look at factor loading basically to improve my model.
Salman heydari khajehpour It is very uncommon to have a non-significant p-value for the chi-square, especially if you have a complex model or a large sample size. I almost never report that measure. I usually stick with RMSEA, CFI, cmin/df.
I can't remember if your variables are all ordinal, but if not, if they are continuous, then normality affects model fit as well.
AMOS is CB SEM, so it assumes you will covary all exogenous variables, but it is not a requirement to get it to run. The only thing you really sacrifice when you don't covary them is you are unable to get correlation values between the constructs you do not covary.
One other thought on this. If you have multiple DVs, then they are being estimated simultaneously with the other DVs. SPSS doesn't do this, so in AMOS, some of the variance being explained in one of the DVs may actually cannibalize the explained variance in another. In SPSS they are run independently.
Hi, Dr. Gaskin, could you please tell me why the result is different between the analysis done by AMOS and SPSS?
I mean in your amos demo, wc was not significant, but in your spss demo, resource gap something was not significant while wc was significant.
Actually, i usually build a structural model based on multiple regression analyses. but in some model, i found that the regression paths that I hypothesized are not significant even though they were significant in regression analyses with SPSS.
Hello James! Your videos are really helpful. I have a question, maybe you could help. I’m gonna use spss for my thesis analysis. Could I use the regression presented in this video to analyze the influence of independent variables (UTAUT Model) on behavioral intention to adopt a technology?
Thank you
Yes
How to get .sav file because I have .csv file and it is not loading in AMOS. Please help.
.sav is from SPSS. If you don't have that, .csv should still work. You just need to go to File, Data Files, and then click on File Name, and then change the file type to csv (the dropdown just above the open and cancel buttons).
@@Gaskination Thank You Sir.
How do you export the SEM to a table
James, thanks for the great video. I have a simple question. Do I check the regression weights table for the size of the effect or the standardized regression weights table ? because the estimates are different in these tables
The unstandardized values are the actual slope or coefficient for the regression equation. The standardized values always fall between -1 and 1 and so they are easy to compare across multiple paths and multiple studies. If you are in the social sciences, use the standardized effects. If you are in the hard sciences, use the unstandardized effects. I'm in Information Systems (a social science), so I use the standardized effects.
James Gaskin YOu are great ! just great ! so i will use the standardized regression weights table which are also same with the numbers that i get on the model. James last question please... Did you run path analysis or simple regression in this video ?
best regards,
emrah8882002 In AMOS I ran path analysis. In SPSS I ran simple regression.
Amos allows for structural equation modeling. SPSS does a ton of other stuff, but not SEM.
Hi. thanks for the helpful video. I am very new to statistics so I'm sorry if this is a stupid question... would this regression you have done in the video count as Structural Equation Modelling?
In AMOS, yes. But in SPSS, no.