* Encoding: UTF-8.
get file='D:\data\Seminars\nhanes2f.sav'.

* question set 1.
* 1 Get the means for the variables "iron" and "hgb"; 
    include the weighted (population) and unweight size and the 95% confidence intervals
    in the output.

csdescriptives
  /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
  /summary variables = iron hgb
  /mean
  /statistics se count popsize cin(95).

* 2 Get a frequency table for the variable "sex"; include both the weighted (population) 
    and unweighted size and the design effect.

cstabulate
    /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
    /tables variables = sex
    /cells popsize 
    /statistics count deff.

* 3 Is there a relationship between the variables "sex" and "agegrp"?.
cstabulate
      /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
      /tables variables = sex by agegrp
      /cells popsize tablepct
      /statistics count expected
      /test independence.
* no, the p-value is 0.29, which is not statistically significant.

* question set 2.

* 1 Run a t-test with the variable "heartatk as the predictor and the variable "vitaminc"
    as the outcome.
csglm vitaminc by heartatk
      /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
      /model heartatk
      /statistics parameter cinterval ttest
      /test type = adjf.

* 2 Is the result statistically significant?.
* no - the p-value is 0.181.

* 3 Run a regression model with the variable "vitaminc" as the outcome and one 
    continuous and one categorical variable of your choice  Include the regression 
    coefficients, test statistics and 95% confidence intervals in the output.
csglm vitaminc by heartatk with age
      /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
      /model heartatk age
      /statistics parameter cinterval ttest
      /test type = adjf.

* 4 How are the degrees of freedom for the model calculated?.
* the numerator df is one because there is only one predictor.
* the denominator df is the number of clusters minus the number of strata: 62 - 31 = 31.

* 5 Rerun your regression with data from females only.
csglm vitaminc by heartatk with age
      /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
      /model heartatk age
      /statistics parameter cinterval ttest
      /test type = adjf
      /domain variable = female(1).

* question set 3.

* 1 Create a three-level variable from the variable "iron"; try to keep the number of 
    unweighted cases similar across the levels.

examine variables = iron.
* min = 16, mean = 99.45, max = 321.
* no missing data.
graph
    /histogram = iron.
* cut at 85 and 110.
compute iron3 = 2.
if iron lt 85 iron3 = 1.
if iron gt 110 iron3 = 3.
freq var = iron3.

* 2 Run an ordinal logistic regression with this new variable as the outcome and three
    predictors of your choice.
csordinal iron3(ascending) with age female by region
      /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
      /statistics parameter exp cinterval ttest
      /test type = adjf.

* 3 Test the parallel line assumption.
csordinal health(ascending) by female region with age
       /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
       /model female region age
       /statistics parameter exp cinterval ttest
       /nonparallel test.   

* 4 What would you do if the assumption was not met?.
csordinal health(ascending) by female region with age
       /plan file = 'D:\data\Seminars\nhanes2f_plan.csaplan'
       /model female region age
       /statistics parameter exp cinterval ttest
       /nonparallel test parameter.