Simple Interactive Statistical Analysis
Data weighting and corrected statistics in SISA. |
A number of procedures in SISA allow for the analysis of weighted data. It concerns descriptive -univariate- procedures. All procedures provide weighted estimates, for example for means, proportions and confidence intervals. Also the weighting corrected variance is calculated by estimating design effects. The corrected variance in turn allows for the calculation of corrected statistics for weighted data and correct tests of statistical significance. These statistical tests in all cases concern the likelihood of the difference between groups being caused by chance.
For the procedures to work you need to start with basic, raw, one case/person per row individual level data with in the first column the grouping variable, in the second column the dependent continuous or proportion variable, and in the third column the case weight. For variance correction SISA uses a basic procedure by Kish which only corrects for the effects of weighing. It is discussed in this paper. This procedure is adequate for most survey and community designs were basic single stage stratification and non response correction for representativeness is applied. For more complex multi stage cluster sampling or mixed sampling designs this procedure is certainly not adequate and should not be used. Also, the SISA procedures are meant for descriptive univariate analysis and not for analytic multivariate analysis such as multiple (logistic) regression or ANOVA. In these more complex situations it seems best to use one of the dedicated packages to do the calculations such as SPSS complex samples (http://www.spss.com/complex_samples/), epi info complex samples (www.cdc.gov/EpiInfo/), the module “survey” in “R” (www.r-project.org), Wesvar (http://www.westat.com/westat/expertise/information_systems/wesvar/index.cfm) or AM (http://am.air.org/). Most of these packages are freely available.
Procedures.
Frequency. The usual one dimensional frequency table in both unweighted and weighted presentation. Furthermore: weighted means, weighting corrected variance, weighted confidence interval, and the designeffect are presented.
For data input go here.
Help page here.
Beta. RxC Table. RxC table concerns a basic two dimensional crosstable procedure. The procedure matches the values of two variables and counts the number of occasions that pairs of values occur. It then presents the result in a table. Weighted counts and proportions, weighting corrected chi square, weighted chi square, and the designeffect are presented.
For data input go here.
Help page here.
T-Test. The t-test procedure is used to test the likelihood of the difference between two groups being caused by chance. For example, is the difference in average BMI between a sample of males and a sample of females due to “just” having a bit more heavy males in the sample? T-test can be used on both continuous dependent outcomes and dichotomous percentage outcome variables, such as the difference between males and females in the percentage of individuals being overweight. The available procedure can further be used to make a weighted pairwise comparison and a weighted comparison between an estimated sample parameter and a population or historical parameter, and to calculate a weighted NNT.
For data input go here.
Data input help here.
Procedure description here.
Means and Oneway.
The means and oneway procedures are used to study the difference between three or more groups. It concerns continuous mean or average outcome variables only. A weighted means table, F-test for a weighted difference between groups, and weighted multiple comparisons between sets of two means by way of t-tests are available.
For data input go here.
Data input help here.
Oneway procedure description here.
2xr.
The 2xr procedure also studies the difference between three or more groups only now it concerns dichotomous proportion or percentage outcome variables only. Weighted cross tables, Chi-square tests, and weighted multiple comparisons between sets of two proportions by way of t-tests are available.
For data input go here.
Data input help here.
Procedure description here.