I'd say independent. It's different people: that's the key element. They just happen to be in the same line of work - but that's what you're studying, after all.

I'd say independent. It's different people: that's the key element. They just happen to be in the same line of work - but that's what you're studying, after all.

I agree they are independent samples because per definition results dependent sample should be affected by previous sample. In your case the results of sample 1 didn't affect the results of sample 2 in any way.

This is a very interesting issue. Thank you!

I would be much interested in reading on your results.

My only question would be if you can really compare at all... Are you certain that the work load, that the worker's activities and safety mesures haven't changed at all, during 10 years? My experience as a specialist in occupational medicine tells me that , fortunately, many job descriptions and human tasks evolve dramatically in ten years, even if performed by the same individuals, taking into account the experiences gained. Can you compare with fidelity?

Yuting, to answer your question, you should determine if job-types of employees responding to questionnaire can be matched. For instance, you might have a different person working at Station A, performing a specific task. The individual may be different, but the work station (type of...) may be more or less constant. You may have a way to assess the risk level of each individuals work station, which may effect their responses overall. You would want to control for that.

It depends. First, it depends on what you mean. If by dependent you mean does knowing the attitudes in 2004 give you some information to guess the attitudes in 2014, then they are dependent. You probably meant independent and dependent in the sense of which t-test. Now it depends on the structure of your data. If your data are just individuals within one company, then they may be independent. But, say you have sampled both in 2004 and 2014 people from 20 companies. Now you have, in some sense, a dependency because knowing that people from company A were safe in 2004, and knowing a person is in company A in 2014, means that this person may be expected to have safer attitudes than someone from Company B. In this case you would probably use multilevel modeling.

This is a good question, more information is necessary before an answer can be given.

The two samples are statistically independent. This does not rule out relationships between the deterministic parts of the data.  And, as others have pointed out, there may be other difficulties in making comparisons.

The two samples in your study are independent  samples.

The two samples are Independent in your data.

the two samples are independent. thanks

Thanks a lot, professors! 

for the two samples, one is 1000, the other is 400. I did find 10 workers filled in the survey in 2004 and 2014. so, if I use independent t test, do I need to exclude these ten cases? these 10 workers do make the two samples dependent, right?

about whether the two samples can be compared or not, we do have the demographic information of the workers. and the distributions for the two samples are similar. I wonder if this can be used as an reason for which they can be compared?

I read your latest questions about the 10 workers that filled the survey in the two years of interest (2004 and 2014). My opinion is that although that this is a very small number of responses in order of your sample sizes, is better to exclude them for your study, because it can give you some bias.

About the demographic information that is similar between the two samples, it’s not a reason to compare them. You must be clear about the underlying reason to try to compare the safety attitudes of both populations and the knowledge of all the information that you can collect could be useful to explain the results of the statistical test that you will apply over the response that you measured.

Good luck.

 I also think that 10 repeated workers must be excluded while considering them as independent in data analysis.

Hi Federico, Thanks!!

I went to mining sites last week. we have got approved from the supervisor before we went to the sites.

however, on one site, the supervisor said he didn't know we will come to do the survey, although the workers and the supervisor all finished the surveys.  but I can observe that some workers didn't do the survey carefully/seriously. some of them even didn't read the questions and then choose the answer.

do you know how to check this kind of bias which is from the survey process? specifically without the support of the supervisor. 

Thanks!

Hi Yuting,

I think that you have a serious problem.  One way to correct inconsistent responses from a questionnaire is through the instrument itself with some crossed questions that will control if some answers are correct or were randomly answered. You must have requirements for good quality control. If inconsistencies are detected data imputation techniques may apply (which are used when you have lost, incomplete or invalid data).

If the problem is that the workers didn't follow the basic guidelines for filling the instrument, can not be known with certainty what responses are valid or not, which can only be avoided by having good quality control systems and well-defined requirements to meet the survey. If during the process you realize that there was carelessness or lack of seriousness to respond, it is best to completely remove these questionnaires (or repeat them if it is possible), but this applies to those cases in which the problem was detected, but there isn't a method to know the true extent of the inconsistency or responses that were selected at random by the worker.

Good luck!

independent variable