data sample: 206
dependent - continuous sample. normaly distributed
independents - sample sizes are are different between categories.
do i use parametric or non-parametric to analyze?
Dear Eekim,
What is your hypothesis?
But, as a general rule, if your dependent variable is normaly distributed you should use parametric tests. Be aware, type of your hypothesis really matters.
Just an FYI, you can always use a nonparametric (distribution free) test for any data set, either parametric or nonparametric, as long as you recognize you'll experience a bit of power loss or need a slightly larger n to get an equal power. Marascuilo and McSweeney's old text ("Nonparametric and Distribution Free Methods for the Social Sciences", Brooks Cole, California) listed a table of asymptotic relative efficiency (see Table 4-2, page 87) showing the loss of efficiency of a non-parametric test compared to its parametric counterpart. The efficiency depends on the nature of the distribution. The text provides a very complete explanation on pages 85-88.
Parametric approach should serve. But the study Hypothesis is important in the choice of statistical analysis.
OK.
If you are looking for relationship, you can use Pearson correlation.
If you want to assess differences between score based on socio-demographic characteristics, you can use ANOVA.
Finally, if you are looking for effect, go for regression.
Hang on a minute. Your dependent variable is the "accuracy of ECG interpretation among doctors". So how are you measuring this?
It seems to me that a whole bunch of people are telling you things like Ette Etuk and we haven't yet got a clear idea of your study methods or research question.
Please tell us a little about each.
I believe I would go for the nonparametric test, as physicians' knowledge tends to be skewed (high numbers of sharp people) so their knowledge is certainly not normally distributed. Also the sample characteristics in terms of socioeconomic background, etc., is also not normally distributed. Finally, your exam itself may not have an underlying normal distribution of scores. If you want correlation, you can use a spearman's rho rather than a Pearson's r, they are interpreted the same way. Good luck with your study.
Ronán Michael Conroy Richard E. Haas my dependent variable is scores. i rated each ECG answer as right and wrong and i convert it to a weighted score. the scores came back as normally distributed but my independent variables (sociodemographic characteristics) are not normally distributed.
How did you weight the score? And, more importantly, why?
The number of ECGs coded correctly is counted data, bounded by zero on one side and the number of ECGs on the other.
You can use Poisson regression (with robust standard errors) to express the effects of your predictor variables in terms of incidence rate ratios. The use of an exposure variable in Poisson regression also allows you to have a variable number of ECGs per observer, in case you have incomplete data on some observers.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size