Graphs with multiple variables

I’ve been trying to get some graphs prepared from the Knowledge & Attitudes data from Fall 2012 (K&A 2012). One of our chunks of data was a series of questions given to survey respondents about their views of the immorality and probable harm caused by various sexual situations (I call these the scenarios): each scenario specified an initiator, a recipient (for lack of a better term), and the age (15 or 21) and sex of each. The initiator was described as “starting a sexual relationship” with the recipient. They always engaged in the same activities: kissing and touching each other’s genitals. There are a lot of variables. Here’s one way to look at them:

  • Initiator sex
  • Initiator age (2 levels: 15 or 21 years old)
  • Recipient sex
  • Recipient age (15 or 21)
  • Immorality rating (DV)
  • Harm rating (DV)

To make things slightly more complex, I simplified the survey forms by cutting out some of the potential crossings of the variables: initiator and perpetrator sex are fully crossed (M/M, M/F, F/M, and F/F conditions) but the ages aren’t–15-year-old initiators are always paired with 21-year-old recipients, and vice-versa. For analysis (but possibly not graphing) purposes it’s also important to know that some comparisons happened between subjects (there were two more-or-less randomly-assigned forms) and others within subjects, with the order of presentation (sadly) fixed. In the future I may do this with more rigorous crossing of all the variables.

I could present subsets of variables in graphs, but I’m really interested in getting as many (independent) variables as possible represented in a single chart, not least because I expect higher-order interactions, and only showing a few variables might obscure those or even mislead the viewer. Here’s an initial stab with only some of the variables:

s.h.01This clearly shows the interaction between initiator age and sex on ratings of harm. However, we can’t see any information about the recipient age or sex, and I predict there will be important interactions. Below you see essentially the same graph, but for immorality ratings.

s_i_01

So I took a different tack:

s_h_as_ir_01

Since the ages of the “initiator” and “recipient” were locked into only two possibilities (15 -> 21 or 21 -> 15), I tried the approach above. I think it’s an improvement, though–with so much information–it takes a while to plow through. We can see that all combinations of variables resulted in somewhat high ratings of harm, except for two: teen males initiating a sexual relationship with adult females and the reverse (adult females with teen males). This fits nicely with intuitive understanding of the way people think about adult-minor sex: we see adult females with teenage boys as a benign exception. I have (so far) tried one other approach:

s_h_isa_rs
In this graph, I added lines for continuity, with the “groups” being the specific pairings of initiator sex and age with recipient age–recipient sex was left for the x-axis. In general I like the results, and several interesting comparisons are visually represented. However, there’s no real change from the previous graph. Perhaps I could do something similar but make the “groups” be the age/sex pairings of the initiator and recipient, and this would leave one set of bars (and the line between) hovering below the others: the adult female/teen male line. Of course, this is eight different pairings, and eight lines is probably far too many for such a crowded space. More thought required.

Vignettes

This section of the questionnaire presented each participant with two short stories (vignettes) of sexual interactions that were, if not clearly abusive in all minds, at least very, very messed up, with the male in both situations ignoring important cues that the female was not interested in sex with him. After each vignette, participants rated the level of responsibility each person in the story had for the sexual encounter and the amount of psychological distress each would likely experience as a result of the encounter.

The “Acquaintance” vignette described two colleagues working late; the male pressures the female into sex despite some indications that she’s not that into him. The “Intimate Partner” vignette describes a married couple who have sex after the husband gets (irrationally) jealous and pressures his wife.

I do not currently consider the specific vignette to be a variable per se, because the vignettes are different enough that I don’t think they should be directly compared (at least not for most purposes). This fact implies two separate graphs, such as a two-panel design. The two variables I am interested in are the presence or absence of perpetrator consequences (some participants saw an extra sentence at the end of vignettes that said the male was questioned or went to trial, but did not go to prison) and the sex of each actor in the vignette–participants provided “responsibility” and “psychological distres” ratings for the male and female, separately. Here is my stab at the responsibility ratings (scale: 0-10):
v_resp_1
It seems that (a) women in the acquaintance vignette are perceived as being significantly more responsible for the sex than women in the intimate partner scenario, and (b) the presence (or absence) of a description of consequences to the men doesn’t change responsibility ratings significantly–though there seems to be a nonsignificant trend toward a slight increase in responsibility ratings for both partners when perpetrator consequences are described.

Below is the same graph, but with psychological distress as the dependent variable (scale: 1-5):

v_dst_1

This tells a similar story, though the lines for females are higher than for males (in contrast to their relationship in the previous graph) because this DV is different. Females are seen as being harmed more than males in both situations (perhaps this shows that people, in general, recognize a sexually coercive situation when they see it, even though some may say it’s ambiguous… that’s one supposition, anyway). In addition, harm ratings for males rise significantly when participants see the aftermath–to the males–of the sexual encounter. I think it’s notable that male and female harm ratings seem approximately equal in the two vignettes, when there is no description of consequences to the males. However, with description of consequences the males in the acquaintance scenario get a much higher rating of harm than those in the intimate partner scenario–though in no condition are males seen as experiencing distress as high as that experienced by females.

The differential increase in male distress ratings when consequences are shown might be explained by the different text used for each situation (sadly, due to limitations in data collection abilities, this was not counterbalanced):

Acquaintance Vignette male consequences statementA few days later, Tony finds that Lori has accused him of rape. He eventually goes to court for the accusations, but is not convicted.

Intimate Partner Vignette male consequences statementThe next day David is questioned by the police, who tell him that Katrina claims that he raped her and has filed a restraining order. He spends a few days in jail, but he is not convicted of any offense.

You can probably see several confounds between these: the number of words, the fact that David’s “surpise” experience is descibed in more detail, the fact that Tony goes to court, and–this is my bet on the deciding factor, if any–the fact that David spends a few days in jail. Alas! Confounds. Well, let’s not let those get in the way of another chart, this time with both distress and responsibility–as z-scores (to deal with the differing response scales):

v_dst_rsp_1

Personally, I quite like this plot. However, it has the difficulty of making sense of distress versus responsibility, the former arguably indicating sympathy for the person in question and the latter indicating a lack of sympathy. I suppose I could reverse the scales, but that’s perhaps even weirder. Maybe I should just use a 4-panel solution:

v_dst_rsp_2
so, there you have it: adventures in wasting several days of vacation processing data and making graphs. If anyone reads this and wants to comment, such comments and suggestions are, of course, welcome. If anyone nerdier than I wants a look at the code (and a description of the somewhat extensive data processing) for making these things, just email me: drogers then the number 1 at the domain utpa and then edu.

Leave a Reply

Your email address will not be published. Required fields are marked *