|
|
|
|
|
|
by jammygit
2485 days ago
|
|
|
> Our approach had three strengths that set it apart from most of the previous work on the topic. First, we had three waves of data for many of our respondents over a period of two years. I don’t honestly think that 2 years is a meaningful time span to measure how a tech changes people’s lives. It is certainly better than a single snapshot, but some effects take time to manifest - it’s simple behaviourism. More changes require more repetitions. Edit: > Second, we had objective measures of Facebook use, pulled directly from participants’ Facebook accounts, rather than measures based on a person’s self-report. What about twitter, Instagram, and whatever else people use? This study design also cannot show causation, just correlation. The control group is self selecting |
|
|
The study seems pretty sound and they did a multivariate regression for inference too. What is your concern? Confounding factors? Lack of calibration?
> This study design also cannot show causation, just correlation. The control group is self selecting
Sure but then again, causation models are currently barely picking up steam in term of being studied and more of a PhD academia study right now. All you have are statistical inference models that we've been using for most research papers that are doing inference.
While a statistician is going to word conclusion carefully, at the end of the day, somebody is going to have the paint a picture and make some plausible leap.
And then other researchers can build upon the paper and redo it with better data set or a different inference models.