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!
Yes that is fine. Keep in mind though that the low/high is relative rather than absolute. If we were to use an absolute low/high, we would just split it down the middle of the scale (at 4).
Hi James, If I am using a scale with 8 questions (continuous) and computed a mean score for my scale to create a new variable (average mean score for each participant). Can I carry out median split for my new variable?
hello, not sure if you are still answering questions but I hope you are!! for my dissertation I am computing the median split for values ranging from 0 to 6. The median is 3. Do I include the values 3 in the split or not? i.e. do I create the low category with 0-2 values and the high category with 4-6 and ignore those with a value of 3? Thanks x
Yep! Still answering questions for now. Probably a dozen or two each day. As for how to split, it depends on your data. If you are aiming for low and high groups, then you must ask yourself whether 3 is low or high or neither. If neither, then exclude it. If you have a sufficiently large dataset, then just get representative samples from 0-2 and 4-6. If the data is skewed and also smaller sample size, then just include 3 in the smaller of the two sides of the data.
Hi James, this video is very informational. Thanks to you. I just have a query, here you have used a single continuous variable for splitting but, is it possible to split by the median, four/five continuous variable belonging to single latent variable by aggregating those variables. Can you please help me on this.
@@Gaskination Hello Mr Gaskin. As always, you helped me a lot with this video too. Just one more question: I also have a moderator variable that is latent (it has 4 items) and I want to use the average of it (i.e. a new continuous variable as a moderator). I did it and use the median split and worked very well as a moderator, but I still can not find any paper that says that they average the items or anything close to it, but i saw a lot of papers that do multigroups analysis with latent variables as moderators, but they do not say anything about that. Do you know any paper that say that they average the items and then do a median split for the moderator variable? thank you soo much in advance again.
@@agustinvidalbuitano9529 Hair et al 2010 ("Multivariate data analysis") says that there are three ways to represent a latent factor: summative (or average), proxy, or factor score. A sum and an average operate the exact same way in regression relationships. As for doing the median split, you're just creating low and high values and then conducting multigroup analysis. There are probably so many papers that do this. Alternatively, you can always do an interaction instead.
Hi James. Nice explanation but I've one small doubt. For my thesis, I used 4 attitude based items in 7 point Likert scale and I need to split this through using median split that measures high vs low attitude. How should I do that? because when I run the analysis, I haven't got equal values when I use the median split.
Hi James, I am running a conditional pathway analysis and I am currently trying to decide whether to transform my continuous moderator variable (level of identification) into a dichotomous variable (high vs low identifiers). I have read different things about why this shouldn’t be done mostly because of the loss of information in the process. What would you say is a good argument to transform a continuous variable into a categorical one, i.e. when should or shouldn’t this be done? Thank you very much for your help!
I agree. You lose information richness when you transform from continuous to categorical. The only time I would do it is when I have multiple moderators. This would allow me to do a multigroup (low/high) of an interaction, instead of having to do a three way interaction.
Hello James, Thank you so much for this. I just want to ask you, My objective is "to find out levels of academic entitlement attitude of higher education students. for this I have used 5 point Likert scale having 15 items. Now how can I know how many students' having high academic entitlement attitude, how many low, and how many having moderated attitude.
If there are 15 items on 5-point scale, you could average or sum those scales together to create an overall entitlement attitude variable, and then run it through a rank-split in SPSS: ruclips.net/video/6MKRRmrmmng/видео.html
Thank you, I have just 1 question please. My median is 29 and the cumulative % = 51.7. Do I therefore include my median in the low values, so everything greater is in my high value? Or does 29 go in the high values? The value after 29 is 31 at 58.6 and the values before is 25 at 41.4%.
This is completely up to you. I prefer to split along a theoretical line. So, for example, if I'm splitting age, and I have respondent ages from 18-63, with a median of 24, I might ignore the median and just say that the "young" group is ages 18-25 and the "old" group is 25+. If I have enough sample size, I prefer to exclude the middle group. In this example, I might exclude anyone ages 25-29.
Hello, I have trouble while working on my Master theses. What I want to do is compose a magazine of a certain number of pages which has different distributions of two different types of contents.Lets say, a magazine is of 36 pages, and I want to make a distribution of: high in editorial contents of pages versus low in commercial contents in pages. That means, the 36 pages are split into two types of pages: editorial pages and commercial pages.
Hey, thanks for the video! I have a question: I want to split my data into 'high responder' and 'low responder', but I want to make sure that in each cell, there is the same amount of participants. For example, I want to have 25 low responders in group1condition1, 25 high responders in group1condition1, 25 low responders in group1condition2 etc. Do you have an idea on how to split my variables this way?
Hi I am not sure if you're available to help but I am trying to graph a moderation and was told to do a median split to get the high and low mean and sd of my variables and graph this. How would I go about this? Thanks!
You just need the regression equations for these. If you can get the data split, then you can run the regression for each half of the data. Then you can plug these into Google or any graphing tool. Here is a video showing part of this: ruclips.net/video/3bDZe6D26AU/видео.html. This video is for multigroup moderation, but you can do it just as well with high/low (which is essentially multiple groups).
Hi James, is it appropriate to do median split for a Likert scale from 1 to 7 (not at all to highly.....). In that case if median value is 5, then one group below 5 is low and the above 5 is high. Does it make sense? For example, I have price competitiveness and I used it as moderator. The median value is 5. Thanks.
Hi, thanks for your presentation. Very informative. I need to describe a variable manipulation where a continuous variable was split into 5 grps, for the purpose of inclusion as a MANOVA IV. How the groups were split is unknown, and not relevant to my discussion. However, I’m wondering what to call the split. Clearly it’s not a dichotomisation as it has 5 groups, not 2. Is there a term which is typically used to describe variable splits into >2 groupings? Thank you for your time.
Yes, you can have as many binary variables as you would like. They will only have multicollinearity if they explain the same portion of variance in the outcome variable.
This video was very helpful writing my thesis, but i still have a question to ask. In lots of papers and videos which dealt with moderation effect in AMOS, they mostly divide a variable into two components (i.e. male vs female, high involvement vs low involvement). Due to lack of samples, I'm currenlty trying to use rest of data as middle, after diving variable into high and low. Is it possible to check moderating effect using three groups (i.e low, middle, high)? Just as with using two groups, can i constrain them with similar method? (for example, a1 = b1 = c1 paths)
+James Gaskin So, For instance if i have only one path in my model and named it as a1,b1 and c1, would it be a1 = b1, b1=c1, a1=c1 to constrain the model??.
Hi thanks for the informative video. I have a question, in my dataset I calculated a Reliable Change Index of scores for the outcome measure (to measure clinical significance). Now i want to create 3 groups, by dividing the continuous RCI scores into 1. one group that has an RC score bigger than 1.96 2. second group that has an RC score within 1.96 and -1.96 3. third group that has an RCI score smaller than -1.96 I'm btw only interested in the frequency of participants that are in the 3 groups. So ideally, SPSS would give me an output like (133 participants have RCI bigger than 1.96, 140 participants between 1.96 and -1.96, and 60 participants smaller than -1.96) do you know how I can divide the groups this way (using sort of cut-off points)?
The simplest way would be to just sort that index column in descending order. Then, create a new variable next to it to indicate whether the index is greater than 1.96, between, or less than -1.96. You don't have to type the value into every cell. You can type it into one cell and then copy that cell and then select all cells you want to paste it into, and then paste and it will paste into all selected cells at once.
I want to split nutrient food data (ex. high and low intake of milk). Do I still have to use the median value to split in high and low, or do I have to use the total maximum intake and split this value to get the right split?
Really it depends on what is actually considered "low" and what is actually considered "high". For example, if I'm studying teenagers, and I want to split them into early and late teen, I wouldn't just take the median split. Instead I would determine what is considered "late" teen, and then I would split it there.
I want to split my 7 point likert religiosity scale and categorize responses into high, low and middle group to be used in MGA AMOS. The middle group will not be used for analysis in MGA AMOS. Can I use this same technique on a 7 point likert scale as shown in your video? My median value is 3.33. Do I have to take into consideration the nominal value of the 7 point likert scale in that logically, responses 1 to 3 makes up low group and 5 to 7 makes up the high group? I'm just confused on deciding which method to use.
In the case of an ordinal scale, it is best to select bottom three and top three and then skip the middle value (4). This is best because it actually represents low and high on the scale.
Hello, do i use this method for questions like this or is there any other method for spss? Yes No 4.1 Do you have control for savings? 1 0 4.2 Do you have control for income? 1 0 4.3 Do you have access to money for your own? 1 0 4.4 Do you have control for loans? 1 0 4.5 Do you have control for cash expense? 1 0 Range: 0-5 Score: 0= no control 1-2=low control 3-4=average control 5=high control
So, it looks like the yes/no portion is redundant with the 0-5 portion. To do a split on this, you would perhaps first sum or average these five items, and then do a split it.
Hi James, great video. im using spss and for me I cant open the data files window as you did at 2:48 . Could you please explain how to get that window on spss ? Or a method to group the variable on spss please? thank you!
@@FahadMughal-id8xn Interactions are not split. Instead, multiply the moderator by the independent variable. Here is a video: ruclips.net/video/dpxkFGctKwo/видео.html
Thanks for this! I have recoded my variable this way. My question is: I want to compare the low and high values on how they affect the dependent variable. How do I do that when they are still blended into one variable? (I realise the recoding is supposed to help and there's probably an obvious solution, but I'm not sure how to do that. I have only compared two separate variables before. I'd be very grateful for any tips!).
James Gaskin I tried it! Thank you so much!! Sadly, there was no significant difference between low and high scores in my data, but at least I know how to do this now :)
So high and low are looked at as continuous, but how can I say, how many pages out of 36 I can attribute to article type 1, so that the distribution is high, and how many ages out of 36 I can attribute to article type 2, so that the distribution across 36 pages of this type of article is low. Thank you!
I am trying to split a magazine of 36 pages into two different types of contents: perceived editorial and perceived commercial (normal articles and advertisement) so that normal type of articles will amount HIGH distribution across those 36 pages, and commercial type of article will amount LOW distribution across those 36 pages (like 26 normal articles, and 10 advertisements). But low and high are continuous...so I thought I have to work this around with SPSS..
Once splitting the variable into High/Low how would I then run a correlation between eg, "high knowledge of UK brands" against another variable say "age"
There are many different approaches. Most would say it is better (if you have sufficient sample size) to make low= 1 standard deviation below the mean, and make high= 1 sd above the mean.
Lets say, I have a magazine of 36 pages, part of it is made of one type of articles, and the other part is made of different type of articles. How can I split 36 pages into two types of articles, so that one type or articles has high distribution (continuous variable) in pages and another type of articles has low distribution (continuous variable) in pages.
Do you mean you are trying to separate long and short articles? If so, I still don't understand the problem. Can't you just decide for yourself what you consider to be long, and what you consider to be short? Make this the cutoff. Then split as I do in the video above, using the "all values less than" and "all other values" options.
I want to split an age variable by the median (into low and high age). However I have cases where the age value is the same as the median itself, so what do these cases get categorised as?
If you use the Rank Cases function (ruclips.net/video/6MKRRmrmmng/видео.html) then you can determine what to do with them, but usually they round up I believe.
Lets say you have a variable that you have recoded in SPSS to be a Dichotomous variable such that all negative values become 1 and all positive values become 2. However, the Data you have or you collected only has positive values such that after re-coding in SPSS you will only have 2, can you still use this variable in SEM as an Observed variable since its not equally represented.
Yes, you can impute a weighted composite for it. See my video called: "Imputing Composite Variables in AMOS". Then you can just split that variable as shown in this video. Best of luck!
Correct, only so I could test multigroup effects. I realize it would probably be better to use the continuous variable with full variance in an interaction, but I needed to demonstrate a procedure.
You could do the "recode into different variable" function in the Transform menu. This will allow you to specify new values for old values. For example, you could tell it to give the value of 1 to any old value between 25-30.
Just split it as shown in this video (or use any other theoretically reasonable approach), and then use the new grouping variable as the factoring variable in the ANOVA. For the dependent variables in the ANOVA, just use whichever variables you want to observe differences for for these two groups.
Thank you very much for such a nice tutorial series, I am getting help only from your tutorials to learn about SEM and your tutorials are so good and sufficient to learn everything. Thanks. May i ask one question that can i convert my summated scale (a latent construct with five variables) to use it for Group SEM for my analysis. As you explained in age, that is fine but how i can convert my summated scale ???.Is there any of your tutorial for sorting out that problem?
The median just makes more sense to me since you will have a real splitting point. For most variables, the mean won't actually exist as a real response value (particularly with Likert scales). There are different ways to do it though. If you have enough data, the best way is to take a standard deviation above and below the median.
+James Gaskin Hi, Please when you said "the best way is to take a standard deviation above and below the median" Does it mean, LOW values are < Median-SD and the HIGH values are > Median+SD? If so what about values between Median-SD and Median+SD? If not how to calculate the threshold or "to take a standard deviation above and below the median" if for instance, my Median is 4.55 and my SD .8! Thanks for your help.
+Arsène Vigan Yes, that is what I mean. The values in the middle are just not used. This is sometimes not possible when sample size is low. As for how to do it, just sort the data by that column, and then create a new variable where rows with a value in that column that is low, get a value in your new variable of 0; high get 2 and middle get 1. Then you can use the 0 and 2 as grouping values.
Can I ask why because I conducted a multiple regression but it’s a 2x2 design so the analysis only shows the total sum of each predictor rather than the high/low levels of each predictor - how would I find out whether high levels of one predictor was significantly associated with the DV?
@@zaynahcandy9440 If the the variables are sums, then you can use correlation or regression. If there are only two variables, then correlation and regression will result in nearly the same estimates. If you want to explicitly test low/high, then split the explanatory variable into low and high (as shown in the video), and then do an ANOVA or T-test where the outcome variable (that was not split) is the dependent variable in these analyses and the explanatory variable (that was split) is used to group them.
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!
My crops are watered, my skin is clear. Thank you.
You're still replying to comments....it's amazing 👍 awesome work
Yes that is fine. Keep in mind though that the low/high is relative rather than absolute. If we were to use an absolute low/high, we would just split it down the middle of the scale (at 4).
Hi James, If I am using a scale with 8 questions (continuous) and computed a mean score for my scale to create a new variable (average mean score for each participant). Can I carry out median split for my new variable?
No. The stats tools package plotter is for interaction with continuous variables.
hello, not sure if you are still answering questions but I hope you are!! for my dissertation I am computing the median split for values ranging from 0 to 6. The median is 3. Do I include the values 3 in the split or not? i.e. do I create the low category with 0-2 values and the high category with 4-6 and ignore those with a value of 3? Thanks x
Yep! Still answering questions for now. Probably a dozen or two each day. As for how to split, it depends on your data. If you are aiming for low and high groups, then you must ask yourself whether 3 is low or high or neither. If neither, then exclude it. If you have a sufficiently large dataset, then just get representative samples from 0-2 and 4-6. If the data is skewed and also smaller sample size, then just include 3 in the smaller of the two sides of the data.
Hi James, this video is very informational. Thanks to you. I just have a query, here you have used a single continuous variable for splitting but, is it possible to split by the median, four/five continuous variable belonging to single latent variable by aggregating those variables. Can you please help me on this.
+Sanjit Singh Yes. You can average them or create a factor score for them, and then split them.
@@Gaskination Hello Mr Gaskin. As always, you helped me a lot with this video too. Just one more question: I also have a moderator variable that is latent (it has 4 items) and I want to use the average of it (i.e. a new continuous variable as a moderator). I did it and use the median split and worked very well as a moderator, but I still can not find any paper that says that they average the items or anything close to it, but i saw a lot of papers that do multigroups analysis with latent variables as moderators, but they do not say anything about that. Do you know any paper that say that they average the items and then do a median split for the moderator variable? thank you soo much in advance again.
@@agustinvidalbuitano9529 Hair et al 2010 ("Multivariate data analysis") says that there are three ways to represent a latent factor: summative (or average), proxy, or factor score. A sum and an average operate the exact same way in regression relationships. As for doing the median split, you're just creating low and high values and then conducting multigroup analysis. There are probably so many papers that do this. Alternatively, you can always do an interaction instead.
@@Gaskination Thank you so much Mr Gaskin!!!!
Hi James. Nice explanation but I've one small doubt. For my thesis, I used 4 attitude based items in 7 point Likert scale and I need to split this through using median split that measures high vs low attitude. How should I do that? because when I run the analysis, I haven't got equal values when I use the median split.
Suganthi Aquila you could just sort it and split it down the middle.
James Gaskin Thank you sir
Hi James, I am running a conditional pathway analysis and I am currently trying to decide whether to transform my continuous moderator variable (level of identification) into a dichotomous variable (high vs low identifiers). I have read different things about why this shouldn’t be done mostly because of the loss of information in the process. What would you say is a good argument to transform a continuous variable into a categorical one, i.e. when should or shouldn’t this be done?
Thank you very much for your help!
I agree. You lose information richness when you transform from continuous to categorical. The only time I would do it is when I have multiple moderators. This would allow me to do a multigroup (low/high) of an interaction, instead of having to do a three way interaction.
Hello James, Thank you so much for this. I just want to ask you, My objective is "to find out levels of academic entitlement attitude of higher education students. for this I have used 5 point Likert scale having 15 items. Now how can I know how many students' having high academic entitlement attitude, how many low, and how many having moderated attitude.
If there are 15 items on 5-point scale, you could average or sum those scales together to create an overall entitlement attitude variable, and then run it through a rank-split in SPSS: ruclips.net/video/6MKRRmrmmng/видео.html
Thank you, I have just 1 question please. My median is 29 and the cumulative % = 51.7. Do I therefore include my median in the low values, so everything greater is in my high value? Or does 29 go in the high values? The value after 29 is 31 at 58.6 and the values before is 25 at 41.4%.
This is completely up to you. I prefer to split along a theoretical line. So, for example, if I'm splitting age, and I have respondent ages from 18-63, with a median of 24, I might ignore the median and just say that the "young" group is ages 18-25 and the "old" group is 25+. If I have enough sample size, I prefer to exclude the middle group. In this example, I might exclude anyone ages 25-29.
@@Gaskination You are amazing - thank you, thank you, thank you x 1000000
Hello, I have trouble while working on my Master theses. What I want to do is compose a magazine of a certain number of pages which has different distributions of two different types of contents.Lets say, a magazine is of 36 pages, and I want to make a distribution of: high in editorial contents of pages versus low in commercial contents in pages. That means, the 36 pages are split into two types of pages: editorial pages and commercial pages.
Hey, thanks for the video! I have a question: I want to split my data into 'high responder' and 'low responder', but I want to make sure that in each cell, there is the same amount of participants. For example, I want to have 25 low responders in group1condition1, 25 high responders in group1condition1, 25 low responders in group1condition2 etc. Do you have an idea on how to split my variables this way?
You can use a rank split. This makes a specific number of equal sized groups. Here is a video for it: ruclips.net/video/6MKRRmrmmng/видео.html
@@Gaskination thank you so much!
Hi I am not sure if you're available to help but I am trying to graph a moderation and was told to do a median split to get the high and low mean and sd of my variables and graph this. How would I go about this? Thanks!
You just need the regression equations for these. If you can get the data split, then you can run the regression for each half of the data. Then you can plug these into Google or any graphing tool. Here is a video showing part of this: ruclips.net/video/3bDZe6D26AU/видео.html. This video is for multigroup moderation, but you can do it just as well with high/low (which is essentially multiple groups).
Hi James, is it appropriate to do median split for a Likert scale from 1 to 7 (not at all to highly.....). In that case if median value is 5, then one group below 5 is low and the above 5 is high. Does it make sense? For example, I have price competitiveness and I used it as moderator. The median value is 5. Thanks.
Hi, thanks for your presentation. Very informative. I need to describe a variable manipulation where a continuous variable was split into 5 grps, for the purpose of inclusion as a MANOVA IV. How the groups were split is unknown, and not relevant to my discussion. However, I’m wondering what to call the split. Clearly it’s not a dichotomisation as it has 5 groups, not 2. Is there a term which is typically used to describe variable splits into >2 groupings? Thank you for your time.
Can i have two variable high and low? If yes , can i run regression without having Coliniarity between variable?
Yes, you can have as many binary variables as you would like. They will only have multicollinearity if they explain the same portion of variance in the outcome variable.
Hi. Thanks for this great video. Do you have any references for this median split method? Thanks.
Here are several: scholar.google.com/scholar?hl=en&as_sdt=0%2C45&q=%22median+split%22+%22moderation%22&btnG=
This video was very helpful writing my thesis, but i still have a question to ask.
In lots of papers and videos which dealt with moderation effect in AMOS, they mostly divide a variable into two components (i.e. male vs female, high involvement vs low involvement).
Due to lack of samples, I'm currenlty trying to use rest of data as middle, after diving variable into high and low.
Is it possible to check moderating effect using three groups (i.e low, middle, high)?
Just as with using two groups, can i constrain them with similar method? (for example, a1 = b1 = c1 paths)
+이정호 You can use three parts, but then you must compare them in pairs. You can do A:B, B:C, A:C, or you can do A:BC, B:AC, C:AB.
+James Gaskin So, For instance if i have only one path in my model and named it as a1,b1 and c1, would it be a1 = b1, b1=c1, a1=c1 to constrain the model??.
+이정호 You could do it that way, or you could just name the path the same thing across the three models. This does the same thing.
Hi thanks for the informative video. I have a question, in my dataset I calculated a Reliable Change Index of scores for the outcome measure (to measure clinical significance).
Now i want to create 3 groups, by dividing the continuous RCI scores into
1. one group that has an RC score bigger than 1.96
2. second group that has an RC score within 1.96 and -1.96
3. third group that has an RCI score smaller than -1.96
I'm btw only interested in the frequency of participants that are in the 3 groups.
So ideally, SPSS would give me an output like
(133 participants have RCI bigger than 1.96,
140 participants between 1.96 and -1.96, and 60 participants smaller than -1.96)
do you know how I can divide the groups this way (using sort of cut-off points)?
The simplest way would be to just sort that index column in descending order. Then, create a new variable next to it to indicate whether the index is greater than 1.96, between, or less than -1.96. You don't have to type the value into every cell. You can type it into one cell and then copy that cell and then select all cells you want to paste it into, and then paste and it will paste into all selected cells at once.
thanks for the quick response James, I'll try it like that
I want to split nutrient food data (ex. high and low intake of milk). Do I still have to use the median value to split in high and low, or do I have to use the total maximum intake and split this value to get the right split?
Really it depends on what is actually considered "low" and what is actually considered "high". For example, if I'm studying teenagers, and I want to split them into early and late teen, I wouldn't just take the median split. Instead I would determine what is considered "late" teen, and then I would split it there.
Thanks for your quick answer!
I want to split my 7 point likert religiosity scale and categorize responses into high, low and middle group to be used in MGA AMOS. The middle group will not be used for analysis in MGA AMOS. Can I use this same technique on a 7 point likert scale as shown in your video? My median value is 3.33.
Do I have to take into consideration the nominal value of the 7 point likert scale in that logically, responses 1 to 3 makes up low group and 5 to 7 makes up the high group? I'm just confused on deciding which method to use.
In the case of an ordinal scale, it is best to select bottom three and top three and then skip the middle value (4). This is best because it actually represents low and high on the scale.
Thanks James for your advice !! Appreciate it!
Hello, do i use this method for questions like this or is there any other method for spss?
Yes No
4.1 Do you have control for savings? 1 0
4.2 Do you have control for income? 1 0
4.3 Do you have access to money for your own? 1 0
4.4 Do you have control for loans? 1 0
4.5 Do you have control for cash expense? 1 0
Range: 0-5
Score:
0= no control
1-2=low control
3-4=average control
5=high control
So, it looks like the yes/no portion is redundant with the 0-5 portion. To do a split on this, you would perhaps first sum or average these five items, and then do a split it.
Hi James, great video. im using spss and for me I cant open the data files window as you did at 2:48 . Could you please explain how to get that window on spss ? Or a method to group the variable on spss please?
thank you!
that's because at that point I'm using AMOS, not SPSS.
Hi! James I am doin moderation analysis in amos ..
My question is how can I split my moderating variable
Do I need to split interaction variable?
@@FahadMughal-id8xn Interactions are not split. Instead, multiply the moderator by the independent variable. Here is a video: ruclips.net/video/dpxkFGctKwo/видео.html
Thanks for this! I have recoded my variable this way. My question is: I want to compare the low and high values on how they affect the dependent variable. How do I do that when they are still blended into one variable? (I realise the recoding is supposed to help and there's probably an obvious solution, but I'm not sure how to do that. I have only compared two separate variables before. I'd be very grateful for any tips!).
+Herreguda You could use the high and low values as the factoring variable in an ANOVA or T-test and then use the DV as the DV in that test.
James Gaskin I tried it! Thank you so much!! Sadly, there was no significant difference between low and high scores in my data, but at least I know how to do this now :)
So high and low are looked at as continuous, but how can I say, how many pages out of 36 I can attribute to article type 1, so that the distribution is high, and how many ages out of 36 I can attribute to article type 2, so that the distribution across 36 pages of this type of article is low. Thank you!
I am trying to split a magazine of 36 pages into two different types of contents: perceived editorial and perceived commercial (normal articles and advertisement) so that normal type of articles will amount HIGH distribution across those 36 pages, and commercial type of article will amount LOW distribution across those 36 pages (like 26 normal articles, and 10 advertisements). But low and high are continuous...so I thought I have to work this around with SPSS..
Hi, is this the same thing as defining a cut-off value for the variable?
Perhaps. It is just finding a cut-off for what we consider "high" vs "low"
Once splitting the variable into High/Low how would I then run a correlation between eg, "high knowledge of UK brands" against another variable say "age"
Hello, if I am using a conscientiousness scale, would I still use this method to split the personality trait into two categories (high or low)?
There are many different approaches. Most would say it is better (if you have sufficient sample size) to make low= 1 standard deviation below the mean, and make high= 1 sd above the mean.
Lets say, I have a magazine of 36 pages, part of it is made of one type of articles, and the other part is made of different type of articles. How can I split 36 pages into two types of articles, so that one type or articles has high distribution (continuous variable) in pages and another type of articles has low distribution (continuous variable) in pages.
Hi James very helpful video. I was just wondering do you know how to split two continuous variables into groups?
This video should help: ruclips.net/video/6MKRRmrmmng/видео.html
Do you mean you are trying to separate long and short articles? If so, I still don't understand the problem. Can't you just decide for yourself what you consider to be long, and what you consider to be short? Make this the cutoff. Then split as I do in the video above, using the "all values less than" and "all other values" options.
I want to split an age variable by the median (into low and high age). However I have cases where the age value is the same as the median itself, so what do these cases get categorised as?
If you use the Rank Cases function (ruclips.net/video/6MKRRmrmmng/видео.html) then you can determine what to do with them, but usually they round up I believe.
Thanks!
Hi James, I have a question that i wanted to ask you. it concern this concept of a dichotomous variable. Are you available to assist
Please ask your question. I'm happy to answer.
Lets say you have a variable that you have recoded in SPSS to be a Dichotomous variable such that all negative values become 1 and all positive values become 2.
However, the Data you have or you collected only has positive values such that after re-coding in SPSS you will only have 2, can you still use this variable in SEM as an Observed variable since its not equally represented.
Yes, you can impute a weighted composite for it. See my video called: "Imputing Composite Variables in AMOS". Then you can just split that variable as shown in this video. Best of luck!
Does this mean you are making the continuous variable a dichotomous variable?
Correct, only so I could test multigroup effects. I realize it would probably be better to use the continuous variable with full variance in an interaction, but I needed to demonstrate a procedure.
Not that I know of. I would just describe it literally, "Variable x contains values for five possible group memberships: a, b, c, d, and e."
i want to know how to create group of age. for example 25-30. 31-35. 36-40 etc
You could do the "recode into different variable" function in the Transform menu. This will allow you to specify new values for old values. For example, you could tell it to give the value of 1 to any old value between 25-30.
thank you.
How could I analyse trait anxiety as two separate groups (low and high anxiety) in an ANOVA?
Just split it as shown in this video (or use any other theoretically reasonable approach), and then use the new grouping variable as the factoring variable in the ANOVA. For the dependent variables in the ANOVA, just use whichever variables you want to observe differences for for these two groups.
@@Gaskination okay, thank you
How do we do this with a Likert scale with different subscale items ?
You could do a 'recode into new variables' function in the Transform menu.
And how to split into three - low, average and high values?
follow the instructions in this other video, except do 3 instead of 2: ruclips.net/video/6MKRRmrmmng/видео.html
I love you for this.
Thanks, you really helped me with my thesis =)
Thank you very much for such a nice tutorial series, I am getting help only from your tutorials to learn about SEM and your tutorials are so good and sufficient to learn everything. Thanks. May i ask one question that can i convert my summated scale (a latent construct with five variables) to use it for Group SEM for my analysis. As you explained in age, that is fine but how i can convert my summated scale ???.Is there any of your tutorial for sorting out that problem?
Why use the median and not the mean? Do you have a reference for using the median
The median just makes more sense to me since you will have a real splitting point. For most variables, the mean won't actually exist as a real response value (particularly with Likert scales). There are different ways to do it though. If you have enough data, the best way is to take a standard deviation above and below the median.
+James Gaskin Hi, Please when you said "the best way is to take a standard deviation above and below the median" Does it mean, LOW values are < Median-SD and the HIGH values are > Median+SD? If so what about values between Median-SD and Median+SD? If not how to calculate the threshold or "to take a standard deviation above and below the median" if for instance, my Median is 4.55 and my SD .8! Thanks for your help.
+Arsène Vigan Yes, that is what I mean. The values in the middle are just not used. This is sometimes not possible when sample size is low. As for how to do it, just sort the data by that column, and then create a new variable where rows with a value in that column that is low, get a value in your new variable of 0; high get 2 and middle get 1. Then you can use the 0 and 2 as grouping values.
James Gaskin Thanks you are the best!
Thank you very much. you are the best!
Hi is this basically known as the median split?
yes
Thanks very much for the help!
I'm sorry, I do not understand your question.
is this mean split?
yes
Really helpful!
if you want to run a correlation between these two variables, I would not split them.
Can I ask why because I conducted a multiple regression but it’s a 2x2 design so the analysis only shows the total sum of each predictor rather than the high/low levels of each predictor - how would I find out whether high levels of one predictor was significantly associated with the DV?
@@zaynahcandy9440 If the the variables are sums, then you can use correlation or regression. If there are only two variables, then correlation and regression will result in nearly the same estimates. If you want to explicitly test low/high, then split the explanatory variable into low and high (as shown in the video), and then do an ANOVA or T-test where the outcome variable (that was not split) is the dependent variable in these analyses and the explanatory variable (that was split) is used to group them.
I'm very sorry. I still do not understand the problem. I'm sorry I could not be more help than this. Best of luck!
great thank you!! :)
with syntax:
RANK VARIABLES=yourVariable (A)
/PERCENT
/PRINT=YES
/TIES=MEAN.
RECODE PyourVar (Lowest thru 50=0) (50.01 thru Highest=1) INTO medianSplitVar.
EXECUTE.
it is so best
Thankyou! :-)