Week 4 Discussion
I need at least 500 words Initial Post. 250 words for each question. No Plagiarism. Due in 12 hours. I will also attached the replies of other students once they are available. I need 0.5 page for each reply.
Write a minimum of 250 words for each of the discussion questions below:
A decision model may be descriptive, heuristic, or prescriptive.
- In your judgment, what are some important requirements for making good decision models regardless of whether they are descriptive, heuristic, or prescriptive? How would you check the validity of a model that you have selected?
- Describe the reasons why a manager might use a heuristic decision model instead of a prescriptive model. Use a real life example, and explain your reasons within the context of your example.
In your two replies to classmates, explore how uncertainties may alter the outcomes of their real-life example mode
Lindsay Hyde
Predictive Model Building
COLLAPSE
Top of Form
1. An article I read had a quote by Warren Buffet that I really like: “Risk comes from not knowing what you’re doing” (Sharma, 2019). As a student and a new analyst it spoke to me a lot. This quote covers a lot of common ways to mess up a model: pulling the data incorrectly, not understanding the data, selecting the wrong variables for the model, using the wrong type of model. If that weren’t challenging enough, the textbook estimates that 90% of spreadsheets of 150 rows or more are off by about 5%. Luckily, many of these issues can at least be partially solved through data cleaning and descriptive statistics. Descriptive statistics may seem basic, but putting in the time to clean and study data results in more accurate models.
Descriptive statistics is about understanding data “without drawing any conclusions or inferences”(Evans, 2013). The tests differ when using univariate vs multivariate data. At this point in our education we’ve covered a few different ways to clean data of nulls and check for outliers, so I won’t go in to that. One important descriptive process is to examine the distribution of our variables before using them in a model. Models are, of course, very dependent on correlation. But because of this data needs to be examined for feasibility for correlation studies. Are the variables approximately normally distributed? Are they homoscedastic? Then what do the correlations look like for each variable? Is there a risk of multicollinearity? The majority of a model is built before it ever touches a inferential algorithm.
2. Reading about heuristic models made me think of the AT&T commercial tagline: good enough is not good enough. Turns out, sometimes it is. A heuristic model is a model that finds a good solution, but not necessarily an exact one (Evans 2013). While a heuristic model may seem like a lazy or uninformed type of predictive modeling, there are actually a lot of really good reasons to use a heuristic model. First, heuristic models can be useful when you really just don’t have the data to make an accurate prediction, don’t really need an accurate prediction, or a mixture of both.
For example, I’m currently working on a sort of pre-study on the relationship between the amount of time patients were taking narcotics before surgery to the speed of their recovery and quality of life after surgery. The problem is, all the data we have right now is casually patient-reported data gathered by a surveys sent out by email, and it’s definitely not accurate. I couldn’t get an accurate answer if I tried. What I can do, is clean the data the best I can and say that there’s a correlation and it probably looks like ‘this’. Then the research department can decide if a study to get more accurate data is worth investing in. A heuristic model is best in this case because it’s easy, budget friendly and it tells us what we want to know. Even in healthcare, where heuristic solutions are generally not acceptable (the surgeon AT&T commercial), a heuristic model is useful for generalized decision making.
Evans, J. R. (2013). Statistics, Data Analysis, and Decision Modeling (5th ed.). Upper Saddle River, New Jersey: Pearson Central Pub.
Sharma, H. (2019, February 23). The Guide to Rigorous Descriptive Statistics for Machine Learning and Data Science : Retrieved March 11, 2020, from https://medium.com/@himanshuxd/the-guide-to-rigorous-descriptive-statistics-for-machine-learning-and-data-science-9209f88e4363
Bottom of Form
Shanna Mckinney
Decision Modeling & Risk Analysis
COLLAPSE
Top of Form
I believe that one of the most important requirements in the decision-making process is the quality of the data being analyzed. I work with different datasets every day in my role as an Investment Analyst. The rule of thumb I’ve come to live by is “garbage-in, garbage-out”. If the quality and source of the data is not reliable and as accurate as possible, any analysis you complete regardless of the model you implement, will be compromised.
A general understanding of the elements of the dataset is also important. If you don’t understand what the data represents you could incorrectly interpret the results of your analysis. An understanding of what the elements represent could also aid in selecting the type of modeling that would provide the most accurate interruption of the dataset.
You also need to understand the question you’ve been asked to address. Understanding what you are attempting to prove (or disprove) is a good way to help make the decision on which modeling process would provide the most comprehensive analysis of the dataset. Making a case for why the results of the data being analyzed is important to the overall success of the business is paramount.
One approach to assessing the quality of your model could be to separate the dataset between a test set and training set. The initial analysis is completed on the training set. Those results are then applied to the test set to determine validity. Another approach could be to send the results to the business and asked that they review the data and determine if the results make business sense. “There are many approaches for assessing the quality and characteristics of a data mining model. However, no single comprehensive rule can tell you when a model is good enough, or when you have enough data.” (Testing and Validation 2018)
The heuristic model is used when a dataset is so large that an optimal solution may be too difficult to derive. However, a good solution can be found by excluding some of the elements in the dataset. Heuristic decisions tend to incur greater errors during the decision-making process because all the elements are not included in the analysis. (Gigerenzer 2011) This is appropriate when the goal is to arrive at a “good” enough analysis with the understanding that the results do not reflect an optimal solution. While the prescriptive model is designed to find the ideal result. Every data element and every potential factor that could affect the results are included in the analysis to arrive at the optimal result. (Vigliarolo 2019)
The heuristic model may be the best model to use when the dataset is very large. The size of the dataset could hinder the creation of the prescriptive model making it too complex to build. It may also be ideal if all the elements that could influence the results are unknown. A good example of when the heuristic model might be the best fit is during the launch of a new product. In this case, you may be able to use predictive analysis to forecast potential earnings and staffing needs. However, you may not know how the sales of an existing product will affect the launch of the new product or how a competitor’s product(s) could affect your sales. In this instance a “good” enough analysis can be completed.
Reference
Gigerenzer G. and Gaissmaier W.(n.d. 2011) Heuristic decision making. Retrieved from:
https://www.ncbi.nlm.nih.gov/pubmed/21126183
Testing and Validation (Data Mining). (May 8, 2018). Retrieved from:
https://docs.microsoft.com/en-us/analysis-services/data-mining/testing-and-validation-data-mining?view=asallproducts-allversions
Vigliarolo
B. (April 18, 2019). Prescriptive analytics: A cheat sheet. Retrieved from:
Bottom of Form