Need Stats Help

[scienceguy123]

This is for any stats experts out there:

I can do the number crunching, I just need to know which type of statistical tests I should use.

 

In short, I looked at some data regarding an intervention that is done on people of various ages. I am interested in whether there is a significant difference in the success rate of the intervention between different age groups. Please note that the sample sizes are not constant. I think the test I would require would be for a categorical variable, since there are only two outcomes measured here: success or failure. (Success rate is the number of successes / total number of interventions done per age group)

 

ie

 

Age 1-2 ------------- success rate (n=120)

Age 2-5 ------------- success rate (n=60)

Age 5+ ------------- success rate (n=40)

 

thanks

[switcheroo]

Based on the very little information I have about your project, I would do a logistic regression with success or failure as the dependent variable and age group as the independent variable. You would code each participant as having either succeeded (1) or failed (0), and being in age group 1, 2, or 3. In this way you could also control for other variables that might have affected success/failure (e.g., gender, income, who knows) by just adding them to the list of independent variables in the model.

[orange123]

Based on the very little information I have about your project, I would do a logistic regression with success or failure as the dependent variable and age group as the independent variable. You would code each participant as having either succeeded (1) or failed (0), and being in age group 1, 2, or 3. In this way you could also control for other variables that might have affected success/failure (e.g., gender, income, who knows) by just adding them to the list of independent variables in the model.

 

brings back the mmrs from stats! i like the above advice.

 

op, might i also suggest you look into tukey's honestly significant difference test? aka tukey's hsd. from what i gather, you're simply looking for the difference in succes rates between the groups. so you don't really need to look at the non-successful participants right?

 

ofcourse, i could be wrong, but from what i rmb, tukey's works for determining difference between groups for upto 3 groups and preserves the family wise alpha. so it keeps your error rate in check...or something like that.

[switcheroo]

You can only use Tukey's test on continuous data. The OP's data is categorical (each participant either succeeded or failed; they didn't each get a score on a continuous scale), so you can't use a standard Tukey's here.

[orange123]

You can only use Tukey's test on continuous data. The OP's data is categorical (each participant either succeeded or failed; they didn't each get a score on a continuous scale), so you can't use a standard Tukey's here.

 

Ahhhh *illumined* I stand corrected.

[switcheroo]

Can you tell I spent far too much of my life before med school running statistical test after statistical test? Sigh.

[orange123]

Can you tell I spent far too much of my life before med school running statistical test after statistical test? Sigh.

 

Lol, no worries. I actually find that quite admirable.

[scienceguy123]

hi everyone,

 

so i underestimated the difficulty of doing a logistic regression with 3 age groups. I was hoping someone could guide me as to which stats software to use, or if there is a way to do a logistic regression on the following data manually, or with excel, if someone could offer some advice i'd b really gratedul.

 

here is the summary data, which I want to check for significant difference between groups. I also have the raw data.

 

Procedure A

Groups: # success #failure Totals % Success Rate

Children 1 -2 76 23 99 0.767676768

Children 2-5 36 13 49 0.734693878

Children 5 + 9 11 20 0.45

[A-Stark]

You only really have three data points for a logistic regression, so you won't get anything terribly meaningful from that. Equivalent would be a chi-square test among the three age groups. Try something along those lines. What software packages do you have available?

[cheech10]

If you treat your age groups as discrete groups (which your table already does) rather than a continuous variable such as age, you can analyze this as a 2x3 contingency table and use a chi squared test - much easier than log regression. You can find lots of calculators for this online, such as http://www.physics.csbsju.edu/stats/contingency_NROW_NCOLUMN_form.html from http://www.physics.csbsju.edu/stats/. Here is the output for your analysis:

 

 

data: contingency table

A B C

 

1 76 36 9 121

2 23 13 11 47

 

99 49 20 168

 

 

expected: contingency table

A B C

 

1 71.3 35.3 14.4

2 27.7 13.7 5.60

 

chi-square = 8.41

degrees of freedom = 2

probability = 0.015

[scienceguy123]

hi cheech and other posters,

 

appreciate the advice, adn a chi-square seems simple enough, however i'm not sure if it is telling me what i want to know. I want to learn if Group 2 (Children aged 2-5) or Group 3 (children aged >5) each have a procedural success rate that is significantly different from Group 1 (Children aged 1-2). Also I want to be able to know if group 2 and 3 are significantly different from each other with regards to success rate.

 

When I copied my summary table earlier, its formatting became hard to understand. In case it was unclear, group 1 (children 1-2) had a success rate of 0.77 (76 success 23 failure), group 2 had a success rate of 0.73 (36 success 13 failure) and group 3 had a success rate of 0.45 (9 success 11 failure).

 

thanks again for helping me out with this.

[cheech10]

Well, the contingency table analysis tells you that there is a 1.5% chance that all three groups had equivalent outcomes, but it doesn't tell you which group in particular is different. Given your data, I suspect that group C is significantly different than groups A and B, but to verify this you'd have to test each group individually (A vs B, B vs C, and A vs C). Fortunately, these are simple 2x2 tables, and Fisher's exact test or the standard chi squared test will quickly give you the answer - in fact group C is different than group A and group B at a <0.05 significance level. I'll leave the calculation up to you.

[scienceguy123]

great, thanks that solves my problem!

 

 

apparantly doing a binomial distrubtion analysis can also determine significance, but i'm guessing this is the same kind of analysis as doing a chi-squared test

[switcheroo]

Hate to break up the non-parametric test love over here, but if you can avoid a chi-square test you'd be better off. Non-parametric tests are far less powerful than parametric tests. A logistic regression really is the best option. Very easy to run in SPSS or any other basic statistics software. There are ways to calculate logistic regressions by hand, but if you have any more than like 10 participants, you'll want to slam your head into the wall real fast.

[A-Stark]

Hate to break up the non-parametric test love over here, but if you can avoid a chi-square test you'd be better off. Non-parametric tests are far less powerful than parametric tests. A logistic regression really is the best option. Very easy to run in SPSS or any other basic statistics software. There are ways to calculate logistic regressions by hand, but if you have any more than like 10 participants, you'll want to slam your head into the wall real fast.

 

Hmm, this would only be straightforward if the raw data is formatted in the right way. It would be helpful for calculating things like odds ratios or relative risks, but strictly speaking you only need to fit the model to get the standard errors. If you don't have access to SPSS or another package, you can try free software like R, but it has a steeper learning curve.

[scienceguy123]

so i have access to SPSS. i have entered in my raw data which looks something like this:

 

Patient Name-------Age at time of procedure-------Results of procedure

Patient 1--------------14months------------------success

patient 2----------------75 months----------------failure

---

--

-

 

I was having some trouble analyzing between the age groups I am interested in, and so I re-coded my "Age at time of procedure" variable so that it is 3 different variables "procedure age 1-2", "Procedure age 2-5", "procedure age >5".

 

When I try running a chi-square of these 3 variables versus "Results of Procedure" I don't get any between groups significance values. Could any one help with doing this?

 

 

Also, does anyone have any ideas about how to do the logistic regression for this data to determine significant difference between groups? My data can be re-formated or re-coded as needed.

 

 

Thanks,

[switcheroo]

so i have access to SPSS. i have entered in my raw data which looks something like this:

 

Patient Name-------Age at time of procedure-------Results of procedure

Patient 1--------------14months------------------success

patient 2----------------75 months----------------failure

---

--

-

 

I was having some trouble analyzing between the age groups I am interested in, and so I re-coded my "Age at time of procedure" variable so that it is 3 different variables "procedure age 1-2", "Procedure age 2-5", "procedure age >5".

 

When I try running a chi-square of these 3 variables versus "Results of Procedure" I don't get any between groups significance values. Could any one help with doing this?

 

 

Also, does anyone have any ideas about how to do the logistic regression for this data to determine significant difference between groups? My data can be re-formated or re-coded as needed.

 

 

Thanks,

 

Your syntax in SPSS would be:

 

logistic RESULT with AGE /categorical = AGE. (words in caps are where you'd put in the actual names of your variables)

 

 

In the output, you want to look under Step 1 (enter) for your results. Look for Exp( near the bottom of the output--this is the odds ratio for the given variable. Sig is your p value.

 

 

See here for more help: http://www.ats.ucla.edu/stat/spss/dae/logit.htm