SPSS Linear Regression Complete Tutorial with PhD Professor
- 11 hours on-demand video
- 21 articles
- 17 downloadable resources
- Full lifetime access
- Access on mobile and TV
- Certificate of Completion
Get your team access to 4,000+ top Udemy courses anytime, anywhere.Try Udemy for Business
- SPSS Linear Regression for Business or Dissertation
- Visualizing Linear Regression Results
- Advanced topics like interactions and categorical predictors
- Dealing with data, distribution, and missing data problems
- Deep understanding of the process and meaning of the results
There's a lot here, foundational skills and some relevant decisions about how to move forward.
Attached file is for following along, or use your own data and parallel the actions. Attached file has a few extra variables, but also all the ones pictured in the video.
Effect size estimation is an essential part of planning the right-sized study you need.
Effect size, as part of a power analysis, ensures that you put enough, but not too much, effort and resources into your study. Whether this is proper budgeting of a corporation, or how many weeks it will take to collect data for a dissertation, the importance is obvious.
Consider the usual complexity of topics we do research on. Take the study of depression in psychology as an example. Depression is caused by a myriad of personal and situational factors. When we consider all possible things that contribute, we know that there are probably at least 10 important factors. If all equally important then each would explain about 10% of depression. That seems like a small amount, but anything that explains that much is actually very helpful in the overall effort to treat the problem. If your study tested a relationship between exercise and depression, explaining 10% of depression with exercise would be impressive. That would mean that of all possible predictors (e.g. genetics, experiences, cognitive factors, nutrition, attributions) exercise is one of the important predictors. By default, we tend to think something like this will have a 20% or 30% impact, because we are not thinking of how many other factors there are. Or, because we are invested in our own research topic that we usually care deeply about, or has been commissioned by someone who cares a lot.
Do you see how the above is ironic and counterproductive? It isn't immediately obvious, but the over-confidence of researchers leads them to build a lower-power study. That study fails to find an effect, even though they were right all along and there is a relationship. Believing the relationship is strong can lead to failing to detect it at all.
On the other hand, you could get a huge sample size to ensure you'll find an effect if it is there. But this can be wasteful or misguided as well. If you commission a huge study with 6,000 participants you can be pretty confident you fill find a significant effect, even if it is small. But is it worth it to spend all these resources to explain less than 1% of the variance in a dependent variable? Almost certainly not.
What is the minimum effect you want to detect in this study?
How strong is the effect in similar research?
What size of effect is enough to have real-world consequences?
Below is a reference table. If you would guess that your IV might explain 10% of your DV, then you could use .3 as a good assumption of your correlation/regression coefficient.
From my experience, coefficients above .3 are uncommon in research on people. Coefficients above .5 almost never happen in real-world research. I suggest you assume a smaller relationship than your natural inclination, as over-estimation of the effect size is usually the problem, rather than underestimation.
*unstandardized correlation or regression coefficient (r, B)
Variance Explained is simply the coefficient squared.
Percentage column is Variance Explained converted to % (x100)
Effect Size calculators:
Or try searching things like "effect size calculator" for more options.
Or be more specific like "convert cohen's d to f".
- College math
In-depth modular class - learn only what you need! Includes optional modules for basics, advanced, & emergent problems.
Anyone can follow this step-by-step, end-to-end, in-depth tutorial for linear regression. The modular course covers all the possible pitfalls (well, pretty close), but in optional modules so you don't get bogged down with stuff you don't need.
I've poured all my consulting knowledge into this, even complex problems are covered in tremendous detail. However, you can skip the advanced information you don't need or want.
Checking Normality / Assumptions
Simple Linear Regression
Multiple Linear Regression
Univariate and Multivariate Outliers
Interactions / Moderation
Interpretation of Results
Pre- & Post-Hoc Power Analysis
Other Emergent Issues
- Graduate students completing thesis or dissertation
- Job-seekers interested in adding Linear Regression as a data analysis skill
- Academic researchers
- Psychology students interested in data analysis