In a message dated 8/31/98 1:12:44 PM, hanson@econ.berkeley.edu wrote:
>I just read the research article this press is based on,
I can't find any good indications of direction of causality. Indeed, the
path analysis in Figure 1 explicitly assumes that internet usage
affects followup psychological status and not the other way around.
>which can be found at:
>http://homenet.andrew.cmu.edu/progress/research.html
>
>I find it thoughtful, careful, and persuasive. Contrary
>to speculation on this list, the study gives clear
>indications of direction of causality. Furthermore,
>the study fits with my experience.
The claim for causality is that initial loneliness and depression weren't associated with later internet use. However, depression *was*, it just wasn't significant. A non-significant p value is not evidence for no association; you need to look at confidence intervals, which weren't provided. In any case, this doesn't address the issue of whether *changes* in depression and loneliness affect internet usage.
Another interesting point is that during the course of the study, loneliness and depression *decreased* for the group as a whole. They don't provide p values for this, but going from the normal assumption the decrease in depression is statistically significant (p about 0.05) and the decrease in loneliness would merit further study.
Also during the study, local social network decrease but the distant social network increased. They don't provide the logged values, but the improvement in distant social network looks to be significant as well (just a guesstimate, though). This, however, is what the internet is supposed to do.
So the causality is very much up in the air. The uncontrolled results indicate improved in psychological state amongst people gaining access to the internet but an association between high usage and losses in psychological state.
The proper way to address this, of course, is a randomized control group. Maybe life has just been really good in Pittsburg the last two years. They should have randomly assigned these folks to getting a computer and training or an equivalent amount of money. Then questions like mine wouldn't be an issue.
Another problem:
>Compared with participants who completed only the pretest questionnaire,
>participants who completed both were wealthier ($53,300 versus $43,600
>household income, r=. 20, p < 01), more likely to be adults (74% versus
>55%, r = .16, p < .01), and less lonely (1.98 versus 2.20 on a 5-point
>scale, r=-.13, p < .05).
It's bad when your dropouts are biased for a primary outcome measure.
>I'll share this excerpt from their discussion section:
>
>>On-line friendships are likely to be more limited than friendships
>>supported by physical proximity. On-line friends are less
>>likely than friends developed at school, work, church, or in the
>>neighborhood to be available for help with tangible favors,
>>such as offering small loans, rides, or baby-sitting. Because
>>on-line friends are not embedded in the same day-to-day
>>environment, they will be less likely to understand the context
>>for conversation, making discussion more difficult (Clark,
>>1996) and rendering support less applicable.
This is a plausibility argument, not evidence.