วีดีโอ

Warms my heart to hear this! Thanks for your kind words. I hope it was helpful.

sorry I can't be of much help here... this sounds like some econometric analyses. seasonal analyses are well beyond what multiple regression can handle. sorry.

Thanks for your quick reply!@@loren.toussaint

You would typically do pairwise tests between groups at each time point. Or, you could do pairwise tests between time points WITHIN each group. Or, you could do both. Assumptions of these approaches and type I error inflation implications differ in each case, but those are generally the options. You're right on track!

Thank you for the reply- I'm focussing on the pairwise comparisons within each group since the output of the ANOVA was non-significant for the between factor. Really appreciate the help. What do you mean by the different assumptions though?@@loren.toussaint

@@thatguyvince5703 so you would not want to use the between effect as a gateway to doing t-tests between groups. the overall between effect is an AVERAGED effect that will likely be nonsignificant because at baseline there shouldn't be a difference and this is averaged into the between effects across baseline, post and follow-up. do the between comparisons anyway. if you do the within comparisons. there are assumptions of any statistical test (often normality, homogeneity of variance, etc.). I would review them from a trusted text or website for both independent and dependent t-test as well as mixed anova.

Thanks again for the help. The output from the mixed Anova shows the pairwise comparisions between and within effects already or am I wrong? By the way I have an significant interaction effect of Time x Group while main-effect of time is also significant and main effect of group is not. Would it make sense to explore the main effect of time anyway or should I just focuse on the interaction effect? If I decide to analyze the within-effect of time further which t-test do I use since the same participants are measured but groups are switched? I also think I need to split the data-set in SPSS for the group-variable in order to conduct the t-test. For me the only reasonable answer would be a paired t-test I think to explore at which point in time differences can be found.@@loren.toussaint

@@thatguyvince5703 Oh, sorry. Yes, if you have requested the between and within effects, then they should already be there. I'd use those. If you have an interaction, you should focus on that in your interpretation, not time or betweeen groups effects. Yes, you are correct that you need to look at changes over time WITHIN each group OR differences BETWEEN groups at each time. You can request those tests in syntax I believe or do it by splitting by group.

can you explain more what these variables are and which ones are predictors and which one is the outcome? what is your hypothesis?

You are in the right place. You want multiple correlation and regression. Follow the video I've posted but use fa1, fa2, fb1, fb2, fk1, and fk2 as the predictors to predict each of your outcomes. One model like this will be for each outcome of interest. So, predict bank deposits, then stock market, then whatever else. Separate model for each outcome. This is manageable. You can do this. Stay optimistic! 😁

@@loren.toussaint Never thought you would reply, was searching for your profile all over the social media for asking this, thankyou so much!🥲❤️

I really shouldn't get too involved in your coursework requirements. That is something you should discuss with your professor mentoring your research.

no the variables can have different numbers of response options. that is fine.

@@loren.toussaint Thank you.

if parent education is just one variable, then you can use Pearson correlation with student performance. If you have father's education and mother's education measured separately, then yes multiple correlation/regression would work best.

@@loren.toussaint Sir Thank you so much for your reply its helps me a lot and I fo have a doubt sir. Since father's and mother's education is qualitative variable more precisely ordinal variable, Can I able to use correlation or multiple correlation/regression?

@@aishwaryar1998 that should be just fine, especially if father/mother education has more than 4 categories.

@@loren.toussaint Yes sir I collected the data like Business,govt employees,private employees, industrial works like that sir. Thank you so much for fast reply and clearing my doubt sir❤️. I actually doing college project it helps a lot sir!!

Hi Marshyani, All you would need to do in this case is simply compute each subscale score the same way that I show you how to compute a total score. In other words, you simply have several "total" scores.

You can use all of your productivity variables to predict profitability, but you have to use only one indicator of profitability at a time. In the first case you will have 9 models each with the same predictors and the 9 different outcomes of profitability taken one at a time in different models. Then you will have 9 different models with 5 predictors each. A separate model for each outcome is what is needed. Hope that helps.

You should combine the questions into a scale score using this video here and then see my video on correlation of two scale scores: th-cam.com/video/Kis6IwuqKjc/w-d-xo.html

First, we need to be clear that it is the normality of the residuals that is of interest. There are also several other assumptions of mixed model ANOVA. I found this link helpful: psych.wisc.edu/Brauer/BrauerLab/wp-content/uploads/2014/04/Murrar-Brauer-2018-MM-ANOVA.pdf. If assumptions are not met there are varying degrees of concern, depending on which assumptions are not met and how badly they are violated. I think you're going to find that the search for an exact nonparametric alternative is going to be a relatively difficult one. I might consider looking into generalized estimating equations, if needed. Hope that helps.

@@loren.toussaint thanks a lot, will have a look into this

you can compute average or sum. its just the way to create a scale score from individual items.

@@loren.toussaint Is there a benefit to average over sum or sum over average? For example if I want to check if there is a correlation between a persons occupation and a 10 item questionnaire?

@@SkidMcmarxx generally no difference. However, if there is a small amount of missing data, you might save some cases from being deleted by computing the mean (which is unaffected by the missing data). For instance, a person who answers 9 of 10 questions will have a lower sum score because they have one less item to add to the sum, but if you average the items the average is unaffected by having 9 or 10 responses. If you choose mean, you'll have to choose a cutoff where too much missing data means you can't compute the mean. Often that is determined to be 50% of the items, so anyone with 5 or more responses, you compute the mean, anything less and it is missing. I hope that helps.

@@loren.toussaint Yes, thank you so much.

Typically you would sum or average all items and not divide them by subscales. Hope that helps.

I think you're probably referring to a "doubly multivariate" design and analysis in which case you'd want to report the lambda, F, dfs, and p value (eta squared for effect size). tables of means and sds, or a figure would usually accompany this.

typically you would compute the mean of the items to arrive at a variable score

If there are no missing data, then mean and sum will be equally useful. No advantage to one or the other.

@@loren.toussaint thanks!

Yes, that is correct.

ANOVA significance is the significance of the overall model. Look at the sig (p) value for the individual predictors (X variables) under the "coefficients" table. It has the t value from the t-test next to it. for a moderator model, you must multiply X1 and X2 and then enter X1, X2, and X1*X2 all in the same model and look at the sig of the X1*X2 variable. you must only look at the sig of X1 and X2 when X1*X2 is NOT in the model. Once you add X1*X2 into the model you can only look at sig of X1*X2.

put microfinance loans in as the dependent and education, health, and food consumption in as the independents. That should do it. Best of luck!

Hi Zay, Even if you believe you have recoded correctly, it is possible that due to some other factor the scale is not reliable. usually it is an item or two. I am familiar with the RSE, but I'm not sure why this is happening. However, there are difficulties with reversed items when English scales are used in other cultures.

Loren L. Toussaint hi thanks for your reply! How can I tell which items to get rid of?

Loren L. Toussaint I have recoded the correct variables and it still isn’t working, are my items labelled wrong? My values are 1= strongly disagree 2=Disagree 3=Agree 4= strongly agree I think recoded these for some of the questions as stated by the RSE. I am entering the values as items is that incorrect?

@@hanik8918 sounds like everything is correct. Try getting alpha for just positive worded items (not reversed). Where (what country and language) are you using the scale?

Loren L. Toussaint England, United Kingdom!

Yes, this is exactly how you should use these scores.

@@loren.toussaint thank you!

transforming is something different altogether. this computes a scale score only.

this should work fine. dummy coding will distinguish between say males and female and if the unstandardized coefficient is for instance .75 it means that moving the male (coded 0) to female (coded 1) on X corresponds to a .75 increase in the continuously measured dependent variable. so, females are higher than males. if the coefficient were -.75 then males would be higher than females.

You are welcome! Glad it was useful!!

@@loren.toussaint i was wondering what you would do if the scales are different so for example one is on a 5 point scale while one is at a 7 point scale, this is for a uni work so im sure they added that on purpose

@@laaaaaaaa533 I'd probably standardize the items first, then compute a scale score. There are other methods, but standardization is meant exactly for this purpose--to make two different response scales equal.

@@laaaaaaaa533 yes. Standardize items, then compute scale scores, then do regression analyses. In that order.

@@loren.toussaint thank you soo much really grateful

correct. it can be described in either order. doesn't matter.

at 5:13 in the video simply name the variable something like "total" and then in the numerical expression box you can do either item1+item2+item3 OR you can write sum(item1,item2,item3). Either should work. Be careful using sum if you have missing data. Sum will sum whatever items have data and will ignore if other items have missing data. If you use the first approach and there is missing data on even just one item, you will have a missing value for the total score.

Hi, you should be able to use the video and the scoring instructions you find on the web or in this article: Hamilton M. Development of a rating scale; for primary depressive illness. Br J Soc Clin Psychol. 1967 Dec;6(4):278-96. I don't know the HAM-D as well as other depression measures, but it should be pretty straightforward. That is, sum up the relevant items to obtain a total score. Creating categories is another matter, and you'll likely have to see the article or other articles to find the cutoff scores. To create cut off scores you would need to use "if, then" syntax like below: if(hamd<5)newvar=1. if(hamd>=5 and hamd<10)newvar=2. if(hamd>=10 and hamd<15)newvar=3. if(hamd>=15)newvar=4. You will have to find the right scores to fill into the syntax above. Then paste this into a syntax window and execute the commands. I have a video showing how to run syntax too.

Awesome! Glad you found it useful!!

So glad to hear it! Thanks for your kinds words!!

you'll find my email on www.luther.edu/touslo01. you can email me directly if you like. you can also find me on ResearchGate.

In many cases it does make sense to compute a total score and you would do it just as you have described. Sometimes scales don't have total scores and it is always best to refer to the original article where the scale was first published or refer to the test manual if it is available. Hope this helps.

Loren L. Toussaint thank you so much. Few years ago, I started to write a paprer (that I gave up) and I had to deal with something like that. When I performed PCA analysis for trust scale it always showed 3 dimensions (credibility, integrity and goodwill) and I asked myself if it was right to compute a total score of the (3×4) items and talk about trust ...

I think that would be just fine. Its common to do so.

this sounds exactly like what you need! Good fit of this statistic to your design.

hope to help as much as I can. :-)

correct. not sure what you're referring to regarding KHI2, but your understanding appears to be correct.

its the same as a total score on a test. it represents if the person is high/low on given construct.

Hi loren. Do we have a cutt off score that differentiates between a high/low score? I have an instrument with 6 items and a 5 point likert scale. How to i determine a cutt off score? or if its continuous, as in the person is high/ low on a given construct, is there any specific point (number) to differentiate them?

generally all you need to include is the coefficient alpha value itself and typically alphas of at least .70 are expected. you typically woudn't include the item-total correlations in a formal paper or report. hope that helps.