Week 10: 7.16-7.22
Among this week, I presented my project to 13 students in our program. I found this task very helpful in term of organizing a overall structure for the essay I am writing right now. The presentation was consist of six different parts: background, hypotheses, methods, results, discussion, and limitation, which shared the same structure with my report. By preparing this presentation, I traced through the logic in my background/introduction, rewrote my hypotheses from the one in the proposal, made an additional demographic breakdown analysis for participants, converted SPSS outputs to a written form with well illustrated diagrams, and came up with detailed result discussion. I hoped this presentation could help other students, who were only high schoolers and just started on their project, have a better understanding on research procedures.
There was one more thing happened during the meeting I want to record here. As several students explained their worries about data collection procedures, I remembered that two years ago, my teammates and I used a online survey design tool to collect data. After I mentioned such website exits on the Internet, several students came and asked me for additional information after class. Later that day, I did some research, tracing back to the group chat happened two years ago, and finally found the correct link for the online survey design tool. I, then, sent an email to all participants in that meeting, sharing the tool to people who might find it useful. I really enjoyed doing so, especially when I received a thank-you note from one of the students. I was glad that I could help.
During this week, I also worked on my report writing. Nothing special happened. I just finished Methods and Results parts. Since I had written lots of APA style reports before, it was quite straight forward. In addition to the data analysis I did before, to better serve this report, I ran several frequency and descriptive analyses. To better describe my participants, I also mentions dropped-off data and cited several literatures.
I am going to finish up writing within this week.
Week 9: 7.9-7.15
Last week for data analysis. I finally figured out how to analyze the relationship between emotion regulation ratings and high/low decision making. I was such a fool before!
As I said last time, correlation tests, neither Spearman’s rho nor Pearson’s r worked. Mr. Gu recommended me to run a chi-square or a fisher’s exact test. However, both of them failed as well. Chi-Square is not the right one as it compared the frequencies of choosing A or B for each rating score. (Definitely not the result I am looking for.) I, then, searched on YouTube for SPSS tutorial—relationship between a nominal and a interval variable. YouTube guided me to use Eta test, which failed as well.
When I was struggling about this, I suddenly realized that my mind was preoccupied by the concept to set D_Q1_N and D_Q2 as dependent variables, as my entire data analysis process had always been done in that way, and that was WRONG! The trick is, to analyze the relationship between emotion regulation ratings and high/low decision making, D_Q1-N and D_Q2 should be the independent variables! I am categorizing participants to high-risk and how risk takers, to see how emotion regulation ratings vary. I only need to use independent sample t test and one way ANOVA! That’s Basic Statistics! Easy!
Overall, the result indicates that Mean_H_S is significantly associated with people’s decision making for D_Q2. People with higher self reported happiness suppression ability tend to choose low risk/ low reward option. This result interested me to analyze whether such impact differed in conditions and age groups. However, no significance is revealed.
Now, I am working on the presentation and report. I wish to hear feedbacks from mentors and other students on Thursday.
Week 8: 7.2-7.8
During this week, I refined my data analysis. In order to do so, I first revisited the textbook I read in detail years ago in Statistics class for Psychology, and I, again, watched more on YouTube tutorial videos. Through these, I found some mistakes I made before.
First of all. I changed the analysis model from general linear model to two factor analysis of variance to examine the impact of age and condition on risk-taking decision making. This was because I realized that two factor analysis of variance is more appropriate, since there were only two independent variables in the design. Two factor analysis of variance could analyze the main effect of each independent variables and the interaction of age x condition. The result with D_Q2 as dependent variable indicates a main effect of condition only. Afterward, I ran pairwise comparison to examine which means are different and how much are they different, with Bonferroni correction to correct the problem with multiple comparison. The result indicated that there was a significant difference between happiness condition and sadness condition.
Mr. Gu recommended Linear Regression to better analyze the impact of age. However, I checked on various materials, read that linear regression only apply for scalar variables, or it will be difficult to define x, y, and interception values on the graph. Meanwhile, two factor analysis of variance should be enough to examine the impact of age, as it tested the main effect of age, and interaction between age and condition, which accounts for mean differences between treatment conditions that are not predicted from the overall main effects from both IVs(independent variables). Unfortunately, the result indicated no significant impact of age. Although comparisons between means of each blocks showed that as age increase, participants’ response to emotion became more extreme on decision making, standard deviations were too big to show any significance.
I than sorted the scalar result from D_Q1 to nominal data and ran the same test. Results showed no significance for any of the hypothesis.
I also reran the three-way Chi-Square stats test for independence again. Same results showed as last time. No significance showed for the new data group—nominal D_Q1.
For correlation test, there is something confused me a lot. To examine the relationship between emotion regulation ratings and high/low risk decision making, I ran bivariate Pearson’s r correlation test for scalar data D_Q1, and Spearman’s rho test for nominal D_Q1 and D_Q2. I am not sure if I should use Spearman’s rho for this. As nominal D_Q1 and D_Q2 are all categorical data. Spearman’s rho ranks data to ascending orders, but my categorical data are all about choices, preference, and frequency. Why should them be ranked for correlation test? I think this might not be the right test to run, but I don’t really know what tool to use instead.
I am going to meet with Mr. Gu again, we will try to figure that out by then.
Week 7: 6.25-7.1
Data analysis is hard. I was a straight A student in Statistics in Psychology course. I got top scores in all assignments, papers, and exams in that class. However, doing data analysis for an independent research project is totally different from taking a Stats class in college. I need to not only apply everything I learnt before, but also self-study all the gaps between theories in the textbook and practice in the reality.
The first thing I did was reviewing all the assignments and SPSS files I did for the Stats course and the some parts of the Stats text book. After reviewing my pervious knowledge, I got a sense that the analysis tool I need to use would be General Linear Model, because I had multiple factors and dependent variables to look at, which could not be analyzed by a single t test, or one/two way ANOVA.
To get start with, I imported data from the excel file I made before, and I deleted some of the unnecessary raw data and limited the number of columns to the minimum requirement. Things did not go well in the first. I tried to run a General Linear Model test with multivariable, but the system failed every time and said “certain variables cannot be added as a factor to analyze”. I gave that up and tried to run a Oneway ANOVA with D_Q1(participants’ answer to the 1st question in decision making questionnaire ) as dependent variables and conditions as factor. The output page finally showed up, with result indicated no significance was found in between. I got reaffirmed that one way ANOVA is not the corrected analysis tools my data needed. I still need to go back to GLM and there must be something wrong with my using methods. I looked on YouTube for SPSS’ GLM tutorial video and started learning from the very basic.
Thanks to those tutorials that I found my mistake, which was I did not corrected label my variables in Variable View. I mislabeled variables as Scale and Nominal, and I forgot to transform nominal value to numerical numbers. After fixing that, I got the output I was looking for. GLM showed that there is a significant mean effect of test condition on forced choice decision making task (D_Q2), with F(2)=5.51, P=.005. Compared the forced choice decision making task for each test condition, the pairwise comparisons showed that comparing test condition Every Routing to Happiness, F=.22, with p=.033, and comparing test condition Sadness to Happiness, F=.34, with p=.001. These results revealed that there are significant differences on forced choice decision making task for conditions between ER and H, and between H and S.
In addition, I ran a Chi-Square test for forced choice decision making task (D_Q2), since D_Q2 is an A or B question, with participants choosing whether A or B as answers. I thought an analyzing tool specifying in frequency might be a better choice. The result showed that χ²(2, N=127)=11.51, with p=.004, meaning test conditions have a significant influence on forced choice decision making task.
In order to analyze the relationship between participants’ answers to the emotion regulation questionnaire, I ran correlation tests between each of the emotion regulation ratings (which were divided to four different categories) and answers to decision making (including D_Q1, relative preference rating, and D_Q2, forced choice). The result showed that r(N=127)=-0.216, with p=.015, indicating a negative correlation between happiness suppression skill rating and forced choice decision making task. This means people with higher rating on happiness suppression skill tend to make decision with lower risk. I, then, sorted my data by age and ran Pearson’s correlation test again, but the result indicated no significance between any two sets of data.
After discussing with Mr. Gu, I realized I made tons of mistakes in the process above, or it is better to articulate that as “there are still room of improvement I can make in addition to what I already did.” First, in addition to General Linear Model, I can also run a linear regression to see the impact of age on the relationship between test conditions and high/low risk decision making. Second, for D_Q1, the relative preference ratings for high/low risk decision making questionnaire, instead of labeling the ratings as continuous interval scale, I could relabel them as three different nominal categories (pro-low-risk, pro-high-risk, and medium), and run GLM test again to ameliorate the level of individual differences. Third, in addition to Chi-Square, I can also run the Spearman (I will spend more time to clarify the difference between Chi-Square and Spearman’s r).
I will follow all the suggestions made by Mr. Gu during the coming week and refine my data analysis.
Week 6: 6.18-6.24
I finally finished the entire data collection process. The closer it gets to the end, the hard it will be. During this week, I first carefully recorded all the data I got before to calculate the exact amount of male/female participants required in different conditions. During that process, I realized I made a slight mistake.
There was one thing I did not pay attention to before, which costed me extra amount of time and money to fix it. I did NOT balance for gender in each condition when I gave out the questionnaires. I got much more female samples than male samples. Furthermore, certain blocks had much more female than male sample (7 female, and 3 male). In the proposal, I said I want to have 10 participants in each block (block means a certain condition with a certain age group; I have 12 blocks in total), and I thought that I will balance participants for gender to ensure there are equal number of female and male in each block. Things went off during the process before I realized it. As a result, I discussed with my mentor, Mr. Gu, and decided to recruit 7 more participants to ensure that there will be at least 5 female and 5 male in each block. This is because we want to rule out the effect of gender, as it is possible that male could be more risk taking than female in general or vice versa.
In addition, I got stuck in finding 13-14 year olds. Though I have enough data now, it was not a easy process. First, there was only limited amount of middle schoolers in libraries. Second, I also went to several Middle Schools, but school officers did not allow me to collect data in the school. I ended up asking kids in parks, restaurants, and streets. I grabbed even chance that I encountered with them. For example, once, a 14 year old boy and I were on the same ride with Uber Pool, I asked him to participate as well.
I interviewed 128 people in total, and recorded 127 sets of data. A 25-year-old man misread the instruction on experience sharing task. He was assigned to the happiness condition, but he shared a sad experience first, and then a happy experience, so I removed his data from the pool.
Mr. Gu and I are going to run SPSS software for data analysis. By then, we will have a deeper understanding on the result.
Week 5: 6.11-6.17
I have already collected 84 sets of data. Boston public library is a great place to collect data. There are a great number of college students and grad school students. Meanwhile, there is a teen center in the library that allows me to collect data from teenagers. However, there are still limitations. For instance, there was not enough 19-20 year old sample in the Boston public library, so that I also went to the library of Boston College.
Currently, I am having trouble finding 13 and 14 years old students, and I only got 6 samples in this group. In the teen center, there were more girls than boys, and there was not a lot of 13, 14 years olds. I will try harder on finding them.
I made myself an excel form to record data, and I was very urgent to see the results of the current data set. Though I do not have SPSS software on the my computer, I used the basic calculation tools on excel the calculate means. The result reveals that 16-17 years olds scored similarly on the relatively preference question for risk taking in all three conditions (happiness, sadness, control). However, in the forced choice question for risk taking, teens in the "happiness" condition were much less likely to take the risk than teens in the "control" condition. More detailed data analysis is required to see whether the result is significant or not (probably yes).
19-20 and 23-25 years olds have the same pattern: people in the "happiness" condition scored significantly lower on risk taking task than people in "sadness" and "control" conditions. People acted similarly in "sadness" and "control" conditions. These probably indicate that people aged 19-20 & 23-25 tend to make low risk decisions when feeling happy.
Even though I did not get enough data for 13-14 years olds, I predict that they will mostly choose to play the low risk game. This is because they are less likely to be influenced by the transient emotion induced, and they are more willing to win some money than take the risk.
We still need more samples to analyze what is happening. If things go well, I can finish data collection this week. Let us wait and see.
Week 4: 6.4-6.10
After refining the questionnaires, I asked Mr. Gu to help me print all the materials. Then, I started collecting data. On Wednesday, I came to Boston Public Library and started my data collection. While I was signing my inform consent sheet, I realized that I forgot to add a race/ethnicity question. It will be a problem if I do not ask my participants to answer such question, as all literatures contain a race/ethnicity break-down in the analysis. As a result, I added “race/ethnicity________________” to each copy.
It was very difficult at the beginning, because first, I inaccurately estimate the time required to complete my experiment, and second, this was also my first experience to collect data from random people After asking five people, I finally got the first person who was willing to take the experiment. She was a 22-year-old girl. There were four things I realized from my first sample.
First, the experiment will only take 10 minutes (instead of the 20 minute I estimated before). People are at least 50% more willing to participate when the experiment is about 10 minute.
Second, it is important to leave the participants alone when filling in the surveys, meaning it is inappropriate to stare at the participant when she or he is focusing on the questions.
Third, 22 years old is actually not in my targeted group. The girl was first person who said yes to my experiment, so that I was too exciting to rule out her age restriction.
Four, I asked a question about participants’ education level by using the word “grade”. However, for a 22 years old, it is inappropriate to name a graduate school year as a grade.
These problem repeatedly happened during my following 4 experiments. Because of these, I changed the estimated time from 20 to 10 minutes; I left my participants alone when they are doing the survey; I explained what I meant my asking for grade; and I paid more attention on participants’ age before I gave them the surveys.
I collected 33 sets of data in two days. It is also convenient to collect teen’s data, since there is a teen section in the library. I spent approximately 3.5 hours in the library per day. My policy is I only search for participants once in the same location. This is because I would be unsure about whether I asked the person before when I revisit a place. I know there might be new potential participants coming to the place I visited, but I choose to omit they to save the time required to distinguish the new coming people from the people I asked before.
Data collection process now became standardized for me. I normally spent 15 seconds explain the time required, the context, and the potential rewards of my study to see if the person I asked is interested in the study or not. Around 50% of the people I asked were willing to participate. Then I explained the age limit for my study, and politely asked their age to see whether they were targeted. Around 70% of the people who were willing to participate were also targeted.
Data collection is not a process to rush through. I enjoyed the entire process.
Week 1: 5.14-5.20
During this week, Mr. Gu shared with me the schedule of my research project, and generally introduced the things I need to focus on in the following 3 months. This will be an independent research project. I need to come up with research questions, conduct experiments and data analysis, and finally write an APA style paper. It sounds quite straight forward, but it is definitely not a easy thing to so.
The first task for myself is, after reading assigned meta-analysis on adolescent decision making, finding a specific topic that interested me, and coming up with hypothesis, method, design, significance, and expected results.
The assigned reading contained many different perspectives on adolescent decision making. I first quantitatively searched for related literature to see the current and past research progresses on my most interested topics, delay discounting and context dependency. According to various literatures, delay discounting is a very well-studied concept in this area, and most of the current researches focused on the neuroscience that can explain delay discounting. Researchers also compared it among different group of adolescents, such as teenagers with high suicide rate, substance uses, and certain types of mental disorders. I could not really think about new research questions beyond what has been studies, while I am also not good of neuroscience, so that I shifted my focus.
In terms of context dependency, I was thinking about testing the superposed effects of peer presence and heighten-arousal (since I really love the concept that sometimes, scientific approaches oversimplify real-world situations to single component analysis). However, after reading the designs and methods used in literatures, I felt a experimental setting that test both factors might be quite difficult to set up. Then, I accidentally found a research article discussing the role of emotion on decision making, which then became my topic.
There are only limited researches studied the effects of developmental changes in emotion regulation on adolescence decision making, though the relation between emotion and decision making is well-studied (not throughly) in different groups of adults. In the past, researchers found emotional distress is highly associated with high-risk decision making. For instance, compared with healthy controls, pathological gamblers scored significantly higher on depression, anxiety, and stress scales (Ciccarelli et al., 2016). Another study suggested that different negative emotions have distinguished effects on people's decision making. Results reveals sad individuals favor high-risk/ high-reward options, whereas anxious individuals favor low-risk/ low rewards options (Raghunathan & Pham, 1999).
In terms of adolescence, there are considerable evidence shown adolescence is a period of great activity with changes in brain structure and function, especially in regions coordinate response inhibition, risk and reward evolution, and emotion regulation (Steinberg, 2004). Emotion regulation undergoes the normative developmental processes during adolescence, from heightened emotional arousability, growing regulation of affect, to regulatory competence. As a result, I predict there will be normative changes in adolescence decision making taking in perspective of their distinguished emotion regulatory ability.
Up to this point, I am very interested in studying the effects of emotion and emotion regulation on adolescent decision making, and also the developmental changes associated with this. The experiment could be consisted of three emotion-inducing tasks, self-report emotion regulation, and decision making tasks. We can analyze data with age differences, emotion regulation ability differences and also emotion differences. I am very very excited about this idea, mostly because it looks doable, and I couldn't really find anyone who had done the same thing before.
This project, for me, is not a “task” to complete. It is more likely to be a valuable thing I need to achieve by myself. This will be a time consuming process, but it will be worth it.
