Couples were asked 54 questions related to humiliation, criticism, hard beginnings, common meaning and forgiveness, love maps, negative conflict behaviors, masonry, and defense.
Yöntem and İlhan (2017, 2018) developed the Divorce Predictors Scale based on Gottman couples therapy (Gottman, 2014; Gottman and Gottman, 2012). Using this scale they’ve done divorce prediction in their research paper. Gottman couples therapy explains the causes of divorce based on empirical research.
Yöntem and İlhan (2017, 2018) collected the data from 7 regions of Turkey. In this data out of 170 couples 84 (49.41%) were divorced and 86 (50.59%) were married couples.
We’ll try to gain some insight between happily married and divorced couples.
In this article I’ll be interested in using Divorce Predictors data set to find out the answer to the following questions:
- Is there a significant difference between Happily Married and Divorced couples?
We’ll plot the average values of the 54 questions between Divorced and Married couples.
From the above plot, we can see that there is a difference between Divorced and Married Couples.
𝑁𝑢𝑙𝑙 𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠_𝑖: There is no difference between the divorced and happily married couple for ith question
𝐴𝑙𝑡𝑒𝑟𝑛𝑎𝑡𝑖𝑣𝑒 𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑠𝑖𝑠_𝑖: There is a significant difference between the divorced and happily married couple for ith question
Using the chi-square test of independence with Bonferroni correction for multiple tests with we’ve found the difference between all the questions between the Married and Divorced couples significant.
We can say that there exists a significant difference between happily married and divorced couples.
2. What are the most agreed and disagreed things between the couples?
Using the average difference between married and divorced couples questions we’ve found the top 5 agreed and bottom 5 disagreed questions.
On average most agreed questions between happily married and divorced couples were
- Starting a discussion before what’s going on.
- Humiliating during the discussion.
- Insulting partners during discussions.
- Discussion occurring suddenly.
- Breaking calm during the discussion with their wife. In the top, five disagreed questions on average divorced couples mostly agreed and happily married couples disagreed.
On average most similar opinion topics between happily married and divorced couples were
- They both don’t have time with their partners.
- Acting like strangers who share the same family and environment.
- Staying silent during the discussion being right to hurt their spouse.
- Feeling right in their discussion.
- Most of the time staying silent to calm the environment.
Out of top similar opinion answered on average divorced couples most stayed silent during the conversation, felt more right in their discussion, stayed more silent to hurt their spouse. On the other hand, happily married couples most disagreed with the opinion of sharing home and environment like strangers and not having time for partners.
3. What couples need to focus most to prevent divorce?
Using recursive feature elimination to select the questions that can best predict a divorce. Following 10 questions will be the most important thing couples need to focus on in order to prevent divorce.
1. We don't have time at home as partners.
2. My spouse and I have similar sense of entertainment.
3. My spouse and I have similar ideas about
how roles should be in marriage
4. My spouse and I have similar values in trust.
5. I know my spouse's hopes and wishes.
6. I know my spouse's friends and their social relationships.
7. I can be humiliating when we discuss.
8. We're just starting a discussion before
I know what's going on.
9. I mostly stay silent to calm the environment a little.
10. I feel right in our discussions.
From the above plots and important questions, we can say that they agree with our earlier questions. Divorced couples and happy couples were completely in opposite direction in terms of having a similar sense of entertainments, similar ideas about how roles should be in marriage, having similar values in trust, knowing spouses hopes and wishes, knowing spouses friends and social relationships, humiliating during discussions, starting a discussion before what’s going wrong.
Divorced couples mostly stayed silent to calm the environment, felt right in their discussions while happily married couples stayed almost neutral. In terms of time for partners at home, both agreed that they have time home as partners but divorced couples agree they had less time.
4. How accurately we can predict a future divorce?
Using 70% data as training and 30% data as testing we’ve fitted Logistic Regression, Random Forest, K nearest neighbours, Automated Classifier TPOT. Model training and test accuracy are:
Machine Learning Algorithm Training Test
Logistic Regression 0.974790 0.980392
Random Forest Classifier 1.000000 0.980392
K Nearest Classifier 1.000000 0.980392
Automated Classifier TPOT 0.983193 0.960784
Among the four models, the logistic regression model is the best model as it generalizes well. From this analysis, we can say that we can predict the divorce of couples with 98.03% accuracy.
Our conclusions are
1. There exists a significant difference between married and divorced couples in all the topics of the scale.
2. Divorced and married couple’s top 5 disagreed topics were discussion, humiliating during the discussion, insulting during the discussion, sudden discussion, breaking calm during the discussion.
On the other hand, the top 5 most similar opinionated topics were having time at home for partners, acting like strangers in terms of sharing environment at home. Married couples agreed on average they have more time at home and act less like strangers about family and environment than divorced couples.
The other two topics where married couples were neutral but divorced couples agreed that they mostly stayed silent and felt right in their discussion.
Based on this we can say that married couples are more careful about their discussion, acts less like strangers in terms of sharing environment at home, and gives more time to their partners.
3. Divorced couples and married couples top 10 deciding questions we can see divorced couples and married couples gave opposite answers.
4. Using 70% data as training and 30% data logistic regression model performs best and it can 98.03% accurately predict divorce between couples.
GitHub Repository with python code: https://github.com/ranamahmud/write-a-data-science-blog-post-udacity