Newbie requires help (probably easy for most of you)

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

joakimlindh
Posts: 8
Joined: Tue Jan 27, 2015 2:36 pm

Newbie requires help (probably easy for most of you)

Postby joakimlindh » Tue Jan 27, 2015 3:03 pm

Hey guys, I´m new here, hope you´re all doing well! :)

I´m a 6th year medical student and I´m about to finish with my first ever research. I´ve collected all my data and so on and now it´s time to calculate the magical P-value.

I´m a beginner when it comes to SPSS, but I´ve watched a few youtube tutorials on the subject and I think that I might be on the right track, but still it would be nice to have you guys opinion.

In my research (50 patients), I want to evaluate if there is any correlation between successful treatment and other factors such as for example age.

Here´s my question:
piece of my research.jpg
I´ve made one column, which is headlined Group1 - 1, group2 -2, the value 1 is a representation of a patient with a successfull treatment, and 2 is a representation of a patient with a non-successful treatment. Is this a correct way of putting in the data? If I for example want to investigate the correlation between this column and age, can I simply just choose these two factors into the "bivariate" and get a proper result (pearsons)?
result.jpg
Here´s the result I got, my teacher says that the level is drawn at 0,05. Did I do it right? What does that result tell you guys? Is there a correlation or not? :D My teacher dont wanna bother with me or my research so I´m pretty much on my own here. :(

If any of you would take the time to answer my questions, stupid as they may be, I would be very grateful!

Best regards/ Joakim
You do not have the required permissions to view the files attached to this post.
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Newbie requires help (probably easy for most of you)

Postby GerineL » Tue Jan 27, 2015 3:38 pm

You probably did not use the right analysis: Your "successfull treatment" variable (if i understand correctly: it is binary with 1 = successfull , 2 = unsuccessfull ) is not continuous but nominal.
You should look into t-test, this is probably what you need.
joakimlindh
Posts: 8
Joined: Tue Jan 27, 2015 2:36 pm

Re: Newbie requires help (probably easy for most of you)

Postby joakimlindh » Tue Jan 27, 2015 3:57 pm

GerineL wrote:You probably did not use the right analysis: Your "successfull treatment" variable (if i understand correctly: it is binary with 1 = successfull , 2 = unsuccessfull ) is not continuous but nominal.
You should look into t-test, this is probably what you need.
Thanks alot for your reply GerineL! You have no idea how much time you most probably saved me! :D
Yes, I think you understood me correctly, since it is binary and not continuous. I´m trying to find out if there is any factor influencing the treatment to be successful. The patients that have a "1" in that group reached the target in order to be classified as "successful".

I quickly googled t-test... apparently there are two kinds - independent and dependent/paired sample t-tests. How do I know which one to look into?

Best regards/ Joakim
joakimlindh
Posts: 8
Joined: Tue Jan 27, 2015 2:36 pm

Re: Newbie requires help (probably easy for most of you)

Postby joakimlindh » Tue Jan 27, 2015 7:14 pm

Okay, once again I think I got it now. When I compared column "group1 -1. group2 -2" with age I chose independent t-test och chose group as the independent factor and age as the dependent factor. I found a very similar example on youtube and just copied it.

However, what do I do when comparing group1 -1. group2 -2. with a gender group? Comparing two nomials. Meaning we now have two groups with only two possible answers - 1 and 2? Do I make sense here? :D
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Newbie requires help (probably easy for most of you)

Postby GerineL » Wed Jan 28, 2015 9:44 am

What you do with a t-test, basically, is seeing whether the means in 2 groups differ. In your case, you want to see whether mean age in the "successful" group differs from the unsuccessful group.
You were probably right to choose the independent t-test: You would choose dependent in case you had, for instance, a pre and post measure by the same person (not independent measures: same person), or mother and child health (not independent: same family) etc.

Now if you have 2 nominal variables, you could use chi square, which you can obtain via crosstabs.
This website may be useful to you, you can use this to find out what test you can use for variables with a certain level.

http://www.ats.ucla.edu/stat/mult_pkg/whatstat/

Also, I notice that now you do a bunch of different analyses to figure out how to predict your outcome variable. Instead, you may want to combine all predictors in 1 analysis. In you case, that would be logistic regression, because the thing you want to predict (= dependent variable) is binary.

Good luck!
joakimlindh
Posts: 8
Joined: Tue Jan 27, 2015 2:36 pm

Re: Newbie requires help (probably easy for most of you)

Postby joakimlindh » Wed Jan 28, 2015 10:16 am

GerineL, once again - thanks alot for your advice!

The link you sent me would be very helpful indeed - IF I weren´t such an idiot when it comes to these things... For example, if you have two columns you would like to compare, as in the scenario I described in my earlier post about gender and group1/group2. Isn´t that two independent variables? In the link you provided me with it assumes that you have atleast one dependent variable. Im still not sure which type to choose. :( Sorry for being so thick, but when it comes to statistics... I really, really am a true newbie. :( Please bear with me!
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Newbie requires help (probably easy for most of you)

Postby GerineL » Wed Jan 28, 2015 4:39 pm

no problem! you are right, sometimes questions are more about associations than about a -> b.
however, you still assign one variable dependent, and the other pedictor for most instances.
also, sometimes you might just want to choose the easiest method: if you jsut have 2 variables, it usually doesn't matter (e.g, if you do a regression with 2 vars, it doesn't matter which one is the predictor and which one is the outcome).

Eventhough you may not have a concrete idea about predictor / dependent variable, oftentimes you can still figure out which one should be the dependent an which one should be the independent. For instance, in your example it is quite obvious: You don't expect treatment outcome to cause people to have a certain gender. thus, gender is not the dependent variable here.
joakimlindh
Posts: 8
Joined: Tue Jan 27, 2015 2:36 pm

Re: Newbie requires help (probably easy for most of you)

Postby joakimlindh » Thu Jan 29, 2015 9:34 am

GerineL wrote:For instance, in your example it is quite obvious: You don't expect treatment outcome to cause people to have a certain gender. thus, gender is not the dependent variable here.
Haha! No, you´re right! This makes sense to me! :) Now, finally, I understand the difference between dependent and independent variable! :D But I still don´t understand exactly which type of analysis I´m supposed to use in this case. I´ve been looking through the link you sent me, but it´s akk confusing to me.

You mentioned regression, but there are so many options when I´m looking under that category! :shock:
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Newbie requires help (probably easy for most of you)

Postby GerineL » Thu Jan 29, 2015 10:01 am

Can you give a bit more informtion on what you are trying to achieve?

I think your main goal is to predict treatment outcome based on multiple variables.
You have age, gender ...?

Please for each of the predictors indicate what "level" the variable is: http://onlinestatbook.com/2/introductio ... ement.html

For instance:

gender: coded 0 = girl 1 = boy
Age: Continuous variable (age in years)
...

etc.
joakimlindh
Posts: 8
Joined: Tue Jan 27, 2015 2:36 pm

Re: Newbie requires help (probably easy for most of you)

Postby joakimlindh » Thu Jan 29, 2015 10:35 am

GerineL wrote:Can you give a bit more informtion on what you are trying to achieve?

I think your main goal is to predict treatment outcome based on multiple variables.
You have age, gender ...?

Please for each of the predictors indicate what "level" the variable is: http://onlinestatbook.com/2/introductio ... ement.html

For instance:

gender: coded 0 = girl 1 = boy
Age: Continuous variable (age in years)
...

etc.
Yes, you´re right, or rather IF any of the variables measured affects treatment in any way. And I´m only comparing the "group... successful/not successful..." with one variable at a time. For example: "group+age" "group+type of treatment" etc.

Basically threre are three different types of analysis I need to perform here ("Group" is still the 2-options-group where 1 is successful and 2 is unsuccessful):

1. Group + continous variable (such as age). - This one I managed with T-test (independent)
2. Group + gender or other "yes or no" variables. - This one I need help with. What test to use in order to get a correct P-value. For example in my columns for gender I´ve coded 1 for male and 2 for female.
3. Group + columns with multiple answers, but its not a continuous variable. Im not sure how to explain this one, so I´ll give an example from my research. "Diseases: Diabetic nephropathy - coded 1. Unknown disease - coded 2. Hypertension - coded -3. etc..." In ont column there may be 7 options which are representations of different diseases.

So... I´m guessing from the new link that you sent me that in "analysis no 2" in my list above I need to compare to nominal scales? Hope I atleast got that right, but IF I´m right... what analysis do you suggest that I use?
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Newbie requires help (probably easy for most of you)

Postby GerineL » Thu Jan 29, 2015 3:32 pm

joakimlindh wrote: 2. Group + gender or other "yes or no" variables. - This one I need help with. What test to use in order to get a correct P-value. For example in my columns for gender I´ve coded 1 for male and 2 for female.
You can use chi square here, which you can access via crosstabs. There are a lot of case studies online for this, that might be useful.
3. Group + columns with multiple answers, but its not a continuous variable. Im not sure how to explain this one, so I´ll give an example from my research. "Diseases: Diabetic nephropathy - coded 1. Unknown disease - coded 2. Hypertension - coded -3. etc..." In ont column there may be 7 options which are representations of different diseases.
this sounds like a nominal variable as well (only with more options then your gender variable).
The important question here is: What do you want to know?
You could also use chi square here, which will tell you something like: Is it more likely to have a pos (or neg) treatment outcome with some diseases than others?
joakimlindh
Posts: 8
Joined: Tue Jan 27, 2015 2:36 pm

Re: Newbie requires help (probably easy for most of you)

Postby joakimlindh » Thu Jan 29, 2015 5:12 pm

GerineL wrote:
joakimlindh wrote: 2. Group + gender or other "yes or no" variables. - This one I need help with. What test to use in order to get a correct P-value. For example in my columns for gender I´ve coded 1 for male and 2 for female.
You can use chi square here, which you can access via crosstabs. There are a lot of case studies online for this, that might be useful.
3. Group + columns with multiple answers, but its not a continuous variable. Im not sure how to explain this one, so I´ll give an example from my research. "Diseases: Diabetic nephropathy - coded 1. Unknown disease - coded 2. Hypertension - coded -3. etc..." In ont column there may be 7 options which are representations of different diseases.
this sounds like a nominal variable as well (only with more options then your gender variable).
The important question here is: What do you want to know?
You could also use chi square here, which will tell you something like: Is it more likely to have a pos (or neg) treatment outcome with some diseases than others?
Got it! I found a youtube instruction on how to perform the chi square analysis. My "no2" type analysis with group and gender is now ok - thanks to you! :)

You´re right on the money, I want to know if there if its more likely to have a positive or negative treatment outcome depending on etiology/disease. So.... do I use the chi square test in the same way as with gender and group? :D
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Newbie requires help (probably easy for most of you)

Postby GerineL » Fri Jan 30, 2015 8:51 am

yes!

There are other more complicated analyses possible, but I think this will give you the information you need!!
joakimlindh
Posts: 8
Joined: Tue Jan 27, 2015 2:36 pm

Re: Newbie requires help (probably easy for most of you)

Postby joakimlindh » Fri Jan 30, 2015 3:32 pm

GerineL wrote:yes!

There are other more complicated analyses possible, but I think this will give you the information you need!!
Can´t thank you enough for helping me with this GerineL! :D Finally, I´m able to complete my research work! Take care! :)

Who is online

Users browsing this forum: No registered users and 1 guest

cron