You can always predict either, but unless you collect experimental data, you cannot establish causation except in the rare cases in which you're looking at pre-existing biological conditions that dictate social outcomes. Your hypothesis will dictate your method, so if you predict causation, make sure your method establishes causation rather than a relationship pattern.
I agree with Stephanie that experimental data is the most effective way to determine causality. Sometimes correlational data can provide a strong implication of causality if one variable precedes the other in time (e.g., parents' level of education as a predictor of own level of education). But in most cases, causality is a matter of the hypotheses that you generate, and justify.
I agree with David. One will be looking at causality with experimental data where the hypothesis is to determine whether the presence of factor leads to the other. Correlation will be a weaker way of establishing the same if your hypothesis is to determine whether or not the presence of one thing leads to the other. Having said that a very strong correlation may be suggestive of causality but the appropriate test statistic will have to be used to determine causality.
As has been said, correlational students establish relationships between variables but not cause and effect. For example, there is a relationship between drowning and ice cream consumption. The two factors don't cause each other. It is just that they both increase in warmer weather. There's a correlation between listening to country western music and depression. We is not clear is which comes first. Does the country music make people depressed or do depressed people listen to country music?
On a more serious not this issue is very significant with smoking and lung cancer and other types of cancers. We know there is a relationship between these variables. However it would take an experimental study to establish causality. That's not possible or advisable because of the ethical consideration. It would be unethical to have one group smoke cigarettes and another not and then look at the cancer rate of both groups.
Because the relationships between these two variables are so strong and the number of subjects examined are so many, we can issue warnings about the dangers of cigarette smoking.
Causality shows that one variable causes a achange in the other variable. This is mostly possible in experimental studies where you might have employed some control techniques and be rest assured that ir only the independent variable that caused a change in the dependent variable.
However, in relationship, you only investigate whether a variable goes together or not with another variable; not that one causes a change in the other. Here, you only want to find out if as one variable goes up, the other as well goes up (direct or positive relationship) or as one variable goes up, the other comes down (inverse or negative relationship). It is mostly used in correlational studies to establish relationship between or among variables in a study.
Generally, correlation establishes association but not directionality. Hence, 'A' can be determined to be associated or correlated with 'B', but such a determination does not imply whether 'A' causes 'B'; 'B' causes 'A'; or whether the causation is mutual or reciprocal or bi-directional.
Agree with the previous comments on experimental research.
On the other hand, in survey research, causality can only be established when a temporal dimension is included in the observations; eg with longitudinal data. Causality cannot be established with cross-sectional data.
It is very important that a researcher, especially master or PhD students understand the difference of the two, since this is one of the first questions that the committee would ask, and later, it will determine outcome of a research...
All above mentioned answers are very useful and I will kindly suggest to all of my students to read in details your comments.