During our little break, Valentine’s Day was celebrated. Yet for many, it was a depressing day because they are single and are looking for love.
Speed dating is a popular format (in times of Covid-19 also in virtual form) to meet many different potential soul mates in a short period of time. If you want to learn which factors determine “getting to the next round”, read on!
As you know by now we won’t just speculate and present anecdotal evidence but base our analysis on hard data. In this case, we take a speed dating data set from my colleagues Professor Andrew Gelman (Columbia University) and Professor Jeniffer Hill (New York University) as provided in the Lock5withR package (on CRAN). From the documentation:
Participants were students at Columbia’s graduate and professional schools, recruited by mass email, posted fliers, and fliers handed out by research assistants. Each participant attended one speed dating session, in which they met with each participant of the opposite sex for four minutes. Order and session assignments were randomly determined. After each four-minute “speed date” participants filled out a form rating their date on a scale of 1-10 on various attributes. Only data from the first date in each session is recorded here
The dataset comprises the following 18 feature variables…
LikeMHow much the male likes his partner (1-10 scale)LikeFHow much the female likes her partner (1-10 scale)PartnerYesMMale’s estimate of chance the female wants another date (1-10 scale)PartnerYesFFemale’s estimate of chance the male wants another date (1-10 scale)AgeMMale’s age (in years)AgeFFemales age (in years)RaceMMale’s race: Asian, Black, Caucasian, Latino, or OtherRaceFFemale’s race: Asian, Black, Caucasian, Latino, or OtherAttractiveMMale’s rating of female’s attractiveness (1-10 scale)AttractiveFFemale’s rating of male’s attractiveness (1-10 scale)SincereMMale’s rating of female’s sincerity (1-10 scale)SincereFFemale’s rating of male’s sincerity (1-10 scale)IntelligentMMale’s rating of female’s intelligence (1-10 scale)IntelligentFFemale’s rating of male’s intelligence (1-10 scale)FunMMale’s rating of female as fun (1-10 scale)FunFFemale’s rating of male as fun (1-10 scale)AmbitiousMMale’s rating of female’s ambition (1-10 scale)AmbitiousFFemale’s rating of male’s ambition (1-10 scale)SharedInterestsMMale’s rating of female’s shared interests (1-10 scale)SharedInterestsFFemale’s rating of male’s shared interests (1-10 scale)
…and 2 target variables:
DecisionMaleWould the male like another date? Yes or NoDecisionFemaleWould the female like another date? Yes or No
Without further ado let us fire up our by now well known OneR package (on CRAN) to get right into the matter:
library(OneR)
library(Lock5withR)
data("SpeedDating")
data <- SpeedDating[-c(1, 2)]
OneR(optbin(DecisionMale ~., data = data, method = "infogain"), verbose = TRUE)
## Warning in optbin.data.frame(x = data, method = method, na.omit = na.omit): 76
## instance(s) removed due to missing values
## Warning in OneR.data.frame(optbin(DecisionMale ~ ., data = data, method =
## "infogain"), : data contains unused factor levels
##
## Attribute Accuracy
## 1 * AttractiveM 78%
## 2 LikeM 72%
## 3 PartnerYesM 64%
## 4 SharedInterestsM 63%
## 5 FunM 60.5%
## 6 AmbitiousM 59%
## 7 PartnerYesF 58.5%
## 8 AgeM 57%
## 8 RaceF 57%
## 10 AttractiveF 56.5%
## 10 IntelligentM 56.5%
## 12 IntelligentF 56%
## 13 AmbitiousF 55.5%
## 14 AgeF 55%
## 14 RaceM 55%
## 14 FunF 55%
## 17 LikeF 54.5%
## 17 SincereM 54.5%
## 17 SincereF 54.5%
## 17 SharedInterestsF 54.5%
## 17 DecisionFemale 54.5%
## ---
## Chosen attribute due to accuracy
## and ties method (if applicable): '*'
##
## Call:
## OneR.data.frame(x = optbin(DecisionMale ~ ., data = data, method = "infogain"),
## verbose = TRUE)
##
## Rules:
## If AttractiveM = (1.99,6] then DecisionMale = No
## If AttractiveM = (6,10] then DecisionMale = Yes
##
## Accuracy:
## 156 of 200 instances classified correctly (78%)
We can see that when it comes to men attractiveness is the main predictor. It is even more important than how much he likes her, which comes in second place. The third feature (PartnerYesM Male’s estimate of chance the female wants another date) is particularly interesting:
OneR(optbin(DecisionMale ~ PartnerYesM, data = data, method = "infogain")) ## Warning in optbin.data.frame(x = data, method = method, na.omit = na.omit): 4 ## instance(s) removed due to missing values ## ## Call: ## OneR.data.frame(x = optbin(DecisionMale ~ PartnerYesM, data = data, ## method = "infogain")) ## ## Rules: ## If PartnerYesM = (-0.01,5] then DecisionMale = No ## If PartnerYesM = (5,10] then DecisionMale = Yes ## ## Accuracy: ## 180 of 272 instances classified correctly (66.18%)
It shows that even his expectation of the woman wanting another date with him leads to him wanting another date with her! If you show interest in a potential partner this interest will be reciprocated!
All the other features like intelligence, sincerity, fun, shared interests but also age and race are not very good at predicting the outcome of the date.
Ok, in a way this didn’t come as a surprise, everybody knows that men are superficial beings who only look at the outer qualities of a woman. Surely women will look more at inner values, like intelligence, sincerity, or at least shared interests, right? Let’s have a look:
OneR(optbin(DecisionFemale ~., data = data, method = "infogain"), verbose = TRUE) ## Warning in optbin.data.frame(x = data, method = method, na.omit = na.omit): 76 ## instance(s) removed due to missing values ## Warning in OneR.data.frame(optbin(DecisionFemale ~ ., data = data, method = ## "infogain"), : data contains unused factor levels ## ## Attribute Accuracy ## 1 * AttractiveF 73.5% ## 2 LikeF 68.5% ## 2 FunF 68.5% ## 4 PartnerYesF 66.5% ## 5 SharedInterestsF 66% ## 6 PartnerYesM 62.5% ## 7 AmbitiousM 61.5% ## 8 SincereM 59% ## 8 IntelligentF 59% ## 10 IntelligentM 58.5% ## 10 FunM 58.5% ## 12 LikeM 58% ## 12 AttractiveM 58% ## 12 SharedInterestsM 58% ## 15 AgeF 57.5% ## 16 RaceF 56% ## 17 RaceM 55% ## 18 AgeM 54.5% ## 18 SincereF 54.5% ## 18 AmbitiousF 54.5% ## 18 DecisionMale 54.5% ## --- ## Chosen attribute due to accuracy ## and ties method (if applicable): '*' ## ## Call: ## OneR.data.frame(x = optbin(DecisionFemale ~ ., data = data, method = "infogain"), ## verbose = TRUE) ## ## Rules: ## If AttractiveF = (0.991,6] then DecisionFemale = No ## If AttractiveF = (6,10] then DecisionFemale = Yes ## ## Accuracy: ## 147 of 200 instances classified correctly (73.5%)
Oh dear, the same result. The most important quality to getting to the next level is yet again the attractiveness and the order of the following features is more or less the same, with the notable exception of FunF Female’s rating of male as fun:
OneR(optbin(DecisionFemale ~ FunF, data = data, method = "infogain")) ## Warning in optbin.data.frame(x = data, method = method, na.omit = na.omit): 6 ## instance(s) removed due to missing values ## ## Call: ## OneR.data.frame(x = optbin(DecisionFemale ~ FunF, data = data, ## method = "infogain")) ## ## Rules: ## If FunF = (0.991,6] then DecisionFemale = No ## If FunF = (6,10] then DecisionFemale = Yes ## ## Accuracy: ## 186 of 270 instances classified correctly (68.89%)
So, to summarize: attractiveness is the most important quality, yet you only have limited control over it. But not all is lost if you keep the following in mind: if you are a woman signal that you are interested in the guy, and if you are a guy try to be fun!
If all goes well you should also read my post on the final step: The One Question you should ask your Partner before Marrying! 😉
UPDATE July 13, 2022
I created a video for this post (in German):