Archive for the ‘Probability’ Category
A gentle introduction to Naïve Bayes classification using R
Preamble
One of the key problems of predictive analytics is to classify entities or events based on a knowledge of their attributes. An example: one might want to classify customers into two categories, say, ‘High Value’ or ‘Low Value,’ based on a knowledge of their buying patterns. Another example: to figure out the party allegiances of representatives based on their voting records. And yet another: to predict the species a particular plant or animal specimen based on a list of its characteristics. Incidentally, if you haven’t been there already, it is worth having a look at Kaggle to get an idea of some of the real world classification problems that people tackle using techniques of predictive analytics.
Given the importance of classification-related problems, it is no surprise that analytics tools offer a range of options. My favourite (free!) tool, R, is no exception: it has a plethora of state of the art packages designed to handle a wide range of problems. One of the problems with this diversity of choice is that it is often confusing for beginners to figure out which one to use in a particular situation. Over the next several months, I intend to write up tutorial articles covering many of the common algorithms, with a particular focus on their strengths and weaknesses; explaining where they work well and where they don’t. I’ll kick-off this undertaking with a simple yet surprisingly effective algorithm – the Naïve Bayes classifier.
Just enough theory
I’m going to assume you have R and RStudio installed on your computer. If you need help with this, please follow the instructions here.
To introduce the Naive Bayes algorithm, I will use the HouseVotes84 dataset, which contains US congressional voting records for 1984. The data set is in the mlbench package which is not part of the base R installation. You will therefore need to install it if you don’t have it already. Package installation is a breeze in RStudio – just go to Tools > Install Packages and follow the prompts.
The HouseVotes84 dataset describes how 435 representatives voted – yes (y), no (n) or unknown (NA) – on 16 key issues presented to Congress. The dataset also provides the party affiliation of each representative – democrat or republican.
Let’s begin by exploring the dataset. To do this, we load mlbench, fetch the dataset and get some summary stats on it. (Note: a complete listing of the code in this article can be found here)
It is good to begin by exploring the data visually. To this end, let’s do some bar plots using the basic graphic capabilities of R:
The plots are shown in Figures 1 through 3.
Among other things, such plots give us a feel for the probabilities associated with how representatives from parties tend to vote on specific issues.
The classification problem at hand is to figure out the party affiliation from a knowledge of voting patterns. For simplicity let us assume that there are only 3 issues voted on instead of the 16 in the actual dataset. In concrete terms we wish to answer the question, “what is the probability that a representative is, say, a democrat (D) given that he or she has voted, say, on the three issues?” To keep things simple I’m assuming there are no NA values.
In the notation of conditional probability this can be written as,
(Note: If you need a refresher on conditional probability, check out this post for a simple explanation.)
By Bayes theorem, which I’ve explained at length in this post, this can be recast as,
We’re interested only in relative probabilities of the representative being a democrat or republican because the predicted party affiliation depends only on which of the two probabilities is larger (the actual value of the probability is not important). This being the case, we can factor out any terms that are constant. As it happens, the denominator of the above equation – the probability of a particular voting pattern – is a constant because it depends on the total number of representatives (from both parties) who voted a particular way.
Now, using the chain rule of conditional probability, we can rewrite the numerator as:
Basically, the second term on the left hand side, , is the probability of getting a particular voting pattern (y,n,y) assuming the rep is a Democrat (D). The definition of conditional probability allows us to rewrite this as the probability of getting a n vote for issue v2 and a y vote for issue v3 given that the rep is a Democrat who has voted y on issue v1. Again, this is simply a consequence of the definition of conditional probability.
Another application of the chain rule gives:
Where we have now factored out the n vote on the second issue.
The key assumption of Naïve Bayes is that the conditional probability of each feature given the class is independent of all other features. In mathematical terms this means that,
and
The quantity of interest, the numerator of equation (1) can then be written as:
The assumption of independent conditional probabilities is a drastic one. What it is saying is that the features are completely independent of each other. This is clearly not the case in the situation above: how representatives vote on a particular issue is coloured by their beliefs and values. For example, the conditional probability of voting patterns on socially progressive issues are definitely not independent of each other. However, as we shall see in the next section, the Naïve Bayes assumption works well for this problem as it does in many other situations where we know upfront that it is grossly incorrect.
Another good example of the unreasonable efficacy of Naive Bayes is in spam filtering. In the case of spam, the features are individual words in an email. It is clear that certain word combinations tend to show up consistently in spam – for example, “online”, “meds”, “Viagra” and “pharmacy.” In other words, we know upfront that their occurrences are definitely not independent of each other. Nevertheless, Naïve Bayes based spam detectors which assume mutual independence of features do remarkably well in distinguishing spam from ham.
Why is this so?
To explain why, I return to a point I mentioned earlier: to figure out the affiliation associated with a particular voting pattern (say, v1=y, v2=n,v3=y) one only needs to know which of the two probabilities and is greater than the other. That is, the values of these probabilities are not important in determining the party affiliations.
This hints as to why the independence assumption might not be so quite so idiotic. Since the prediction depends only the on the maximum, the algorithm will get it right even if there are dependencies between feature providing the dependencies do not change which class has the maximum probability (once again, note that only the maximal class is important here, not the value of the maximum).
Yet another reason for the surprising success of Naïve Bayes is that dependencies often cancel out across a large set of features. But, of course, there is no guarantee that this will always happen.
In general, Naïve Bayes algorithms work better for problems involving a large number of discrete features (such as colour, gender or category) even when there are dependencies between features (spam detection is a good example). They work less well f for problems which features are continuous (such as height, weight and age). I’ll say more about this point in a future article in which I will discuss some examples in which Naive Bayes fails.
I hope the above has given you an intuitive feel for how Naïve Bayes algorithms work. I don’t know about you, but my head’s definitely spinning after writing out all that mathematical notation.
It’s time to clear our heads by doing some computation.
Naïve Bayes in action
There are a couple of well-known implementations of Naïve Bayes in R. One of them is the naiveBayes method in the e1071 package and the other is NaiveBayes method in the klaR package. I’ll use the former for no other reason than it seems to be more popular. That said, I have used the latter too and can confirm that it works just as well.
We’ve already loaded and explored the HouseVotes84 dataset. One of the things you may have noticed when summarising the data is that there are a fair number of NA values. Naïve Bayes algorithms typically handle NA values either by ignoring records that contain any NA values or by ignoring just the NA values. These choices are indicated by the value of the variable na.action in the naiveBayes algorithm, which is set to na.omit (to ignore the record) or na.pass (to ignore the value).
Just for fun, we’ll take a different approach. We’ll impute NA values for a given issue and party by looking at how other representatives from the same party voted on the issue. This is very much in keeping with the Bayesian spirit: we infer unknowns based on a justifiable belief – that is, belief based on the evidence.
To do this I write two functions: one to compute the number of NA values for a given issue (vote) and class (party affiliation), and the other to calculate the fraction of yes votes for a given issue (column) and class (party affiliation).
sum_y<-sum(HouseVotes84[,col]==’y’ & HouseVotes84$Class==cls,na.rm = TRUE)
sum_n<-sum(HouseVotes84[,col]==’n’ & HouseVotes84$Class==cls,na.rm = TRUE)
return(sum_y/(sum_y+sum_n))}
Before proceeding, you might want to go back to the data and convince yourself that these values are sensible.
We can now impute the NA values based on the above. We do this by randomly assigning values ( y or n) to NAs, based on the proportion of members of a party who have voted y or n. In practice, we do this by invoking the uniform distribution and setting an NA value to y if the random number returned is less than the probability of a yes vote and to n otherwise. This is not as complicated as it sounds; you should be able to figure the logic out from the code below.
if(sum(is.na(HouseVotes84[,i])>0)) {
c1 <- which(is.na(HouseVotes84[,i])& HouseVotes84$Class==’democrat’,arr.ind = TRUE)
c2 <- which(is.na(HouseVotes84[,i])& HouseVotes84$Class==’republican’,arr.ind = TRUE)
HouseVotes84[c1,i] <-
ifelse(runif(na_by_col_class(i,’democrat’))<p_y_col_class(i,’democrat’),’y’,’n’)
HouseVotes84[c2,i] <-
ifelse(runif(na_by_col_class(i,’republican’))<p_y_col_class(i,’republican’),’y’,’n’)}
Note that the which function filters indices by the criteria specified in the arguments and ifelse is a vectorised conditional function which enables us to apply logical criteria to multiple elements of a vector.
At this point it is a good idea to check that the NAs in each column have been set according to the voting patterns of non-NAs for a given party. You can use the p_y_col_class() function to check that the new probabilities are close to the old ones. You might want to do this before you proceed any further.
The next step is to divide the available data into training and test datasets. The former will be used to train the algorithm and produce a predictive model. The effectiveness of the model will then be tested using the test dataset. There is a great deal of science and art behind the creation of training and testing datasets. An important consideration is that both sets must contain records that are representative of the entire dataset. This can be difficult to do, especially when data is scarce and there are predictors that do not vary too much…or vary wildly for that matter. On the other hand, problems can also arise when there are redundant predictors. Indeed, the much of the art of successful prediction lies in figuring out which predictors are likely to lead to better predictions, an area known as feature selection. However, that’s a topic for another time. Our current dataset does not suffer from any of these complications so we’ll simply divide the it in an 80/20 proportion, assigning the larger number of records to the training set.
Now we’re finally good to build our Naive Bayes model (machine learning folks call this model training rather than model building – and I have to admit, it does sound a lot cooler).
The code to train the model is anticlimactically simple:
Here we’ve invokedthe naiveBayes method from the e1071 package. The first argument uses R’s formula notation.In this notation, the dependent variable (to be predicted) appears on the left hand side of the ~ and the independent variables (predictors or features) are on the right hand side. The dot (.) is simply shorthand for “all variable other than the dependent one.” The second argument is the dataframe that contains the training data. Check out the documentation for the other arguments of naiveBayes; it will take me too far afield to cover them here. Incidentally, you can take a look at the model using the summary() or str() functions, or even just entering the model name in the R console:
Note that I’ve suppressed the output above.
Now that we have a model, we can do some predicting. We do this by feeding our test data into our model and comparing the predicted party affiliations with the known ones. The latter is done via the wonderfully named confusion matrix – a table in which true and predicted values for each of the predicted classes are displayed in a matrix format. This again is just a couple of lines of code:
pred true | democrat | republican |
democrat | 38 | 3 |
republican | 5 | 22 |
The numbers you get will be different because your training/test sets are almost certainly different from mine.
In the confusion matrix (as defined above), the true values are in columns and the predicted values in rows. So, the algorithm has correctly classified 38 out of 43 (i.e. 38+5) Democrats and 22 out of 25 Republicans (i.e. 22+3). That’s pretty decent. However, we need to keep in mind that this could well be quirk of the choice of dataset. To address this, we should get a numerical measure of the efficacy of the algorithm and for different training and testing datasets. A simple measure of efficacy would be the fraction of predictions that the algorithm gets right. For the training/testing set above, this is simply 60/68 (see the confusion matrix above). The simplest way to calculate this in R is:
A natural question to ask at this point is: how good is this prediction. This question cannot be answered with only a single run of the model; we need to do many runs and look at the spread of the results. To do this, we’ll create a function which takes the number of times the model should be run and the training fraction as inputs and spits out a vector containing the proportion of correct predictions for each run. Here’s the function
I’ve not commented the above code as it is essentially a repeat of the steps described earlier. Also, note that I have not made any effort to make the code generic or efficient.
Let’s do 20 runs with the same training fraction (0.8) as before:
[9] 0.9102564 0.9080460 0.9139785 0.9200000 0.9090909 0.9239130 0.9605263 0.9333333
[17] 0.9052632 0.8977273 0.9642857 0.8518519
0.8519 0.9074 0.9170 0.9177 0.9310 0.9643
We see that the outcome of the runs are quite close together, in the 0.85 to 0.95 range with a standard deviation of 0.025. This tells us that Naive Bayes does a pretty decent job with this data.
Wrapping up
I originally intended to cover a few more case studies in this post, a couple of which highlight the shortcomings of the Naive Bayes algorithm. However, I realize that doing so would make this post unreasonably long, so I’ll stop here with a few closing remarks, and a promise to write up the rest of the story in a subsequent post.
To sum up: I have illustrated the use of a popular Naive Bayes implementation in R and attempted to convey an intuition for how the algorithm works. As we have seen, the algorithm works quite well in the example case, despite the violation of the assumption of independent conditional probabilities.
The reason for the unreasonable effectiveness of the algorithm is two-fold. Firstly, the algorithm picks the predicted class based on the largest predicted probability, so ordering is more important than the actual value of the probability. Secondly, in many cases, a bias one way for a particular vote may well be counteracted by a bias the other way for another vote. That is, biases tend to cancel out, particularly if there are a large number of features.
That said, there are many cases in which the algorithm fails miserably – and we’ll look at some of these in a future post. However, despite its well known shortcomings, Naive Bayes is often the first port of call in prediction problems simply because it is easy to set up and is fast compared to many of the iterative algorithms we will explore later in this series of articles.
Endnote
Thanks for reading! If you liked this piece, you might enjoy the other articles in my “Gentle introduction to analytics using R” series. Here are the links:
A gentle introduction to text mining using R
Three types of uncertainty you (probably) overlook
Introduction – uncertainty and decision-making
Managing uncertainty – deciding what to do in the absence of reliable information – is a significant part of project management and many other managerial roles. When put this way, it is clear that managing uncertainty is primarily a decision-making problem. Indeed, as I will discuss shortly, the main difficulties associated with decision-making are related to specific types of uncertainties that we tend to overlook.
Let’s begin by looking at the standard approach to decision-making, which goes as follows:
- Define the decision problem.
- Identify options.
- Develop criteria for rating options.
- Evaluate options against criteria.
- Select the top rated option.
As I have pointed out in this post, the above process is too simplistic for some of the complex, multifaceted decisions that we face in life and at work (switching jobs, buying a house or starting a business venture, for example). In such cases:
- It may be difficult to identify all options.
- It is often impossible to rate options meaningfully because of information asymmetry – we know more about some options than others. For example, when choosing whether or not to switch jobs, we know more about our current situation than the new one.
- Even when ratings are possible, different people will rate options differently – i.e. different people invariably have different preferences for a given outcome. This makes it difficult to reach a consensus.
Regular readers of this blog will know that the points listed above are characteristics of wicked problems. It is fair to say that in recent years, a general awareness of the ubiquity of wicked problems has led to an appreciation of the limits of classical decision theory. (That said, it should be noted that academics have been aware of this for a long time: Horst Rittel’s classic paper on the dilemmas of planning, written in 1973, is a good example. And there are many others that predate it.)
In this post I look into some hard-to-tackle aspects of uncertainty by focusing on the aforementioned shortcomings of classical decision theory. My discussion draws on a paper by Richard Bradley and Mareile Drechsler.
This article is organised as follows: I first present an overview of the standard approach to dealing with uncertainty and discuss its limitations. Following this, I elaborate on three types of uncertainty that are discussed in the paper.
Background – the standard view of uncertainty
The standard approach to tackling uncertainty was articulated by Leonard Savage in his classic text, Foundations of Statistics. Savage’s approach can be summarized as follows:
- Figure out all possible states (outcomes)
- Enumerate actions that are possible
- Figure out the consequences of actions for all possible states.
- Attach a value (aka preference) to each consequence
- Select the course of action that maximizes value (based on an appropriately defined measure, making sure to factor in the likelihood of achieving the desired consequence)
(Note the close parallels between this process and the standard approach to decision-making outlined earlier.)
To keep things concrete it is useful to see how this process would work in a simple real-life example. Bradley and Drechsler quote the following example from Savage’s book that does just that:
…[consider] someone who is cooking an omelet and has already broken five good eggs into a bowl, but is uncertain whether the sixth egg is good or rotten. In deciding whether to break the sixth egg into the bowl containing the first five eggs, to break it into a separate saucer, or to throw it away, the only question this agent has to grapple with is whether the last egg is good or rotten, for she knows both what the consequence of breaking the egg is in each eventuality and how desirable each consequence is. And in general it would seem that for Savage once the agent has settled the question of how probable each state of the world is, she can determine what to do simply by averaging the utilities (Note: utility is basically a mathematical expression of preference or value) of each action’s consequences by the probabilities of the states of the world in which they are realised…
In this example there are two states (egg is good, egg is rotten), three actions (break egg into bowl, break egg into separate saucer to check if it rotten, throw egg away without checking) and three consequences (spoil all eggs, save eggs in bowl and save all eggs if last egg is not rotten, save eggs in bowl and potentially waste last egg). The problem then boils down to figuring out our preferences for the options (in some quantitative way) and the probability of the two states. At first sight, Savage’s approach seems like a reasonable way to deal with uncertainty. However, a closer look reveals major problems.
Problems with the standard approach
Unlike the omelet example, in real life situations it is often difficult to enumerate all possible states or foresee all consequences of an action. Further, even if states and consequences are known, we may not what value to attach to them – that is, we may not be able to determine our preferences for those consequences unambiguously. Even in those situations where we can, our preferences for may be subject to change – witness the not uncommon situation where lottery winners end up wishing they’d never won. The standard prescription works therefore works only in situations where all states, actions and consequences are known – i.e. tame situations, as opposed to wicked ones.
Before going any further, I should mention that Savage was cognisant of the limitations of his approach. He pointed out that it works only in what he called small world situations– i.e. situations in which it is possible to enumerate and evaluate all options. As Bradley and Drechsler put it,
Savage was well aware that not all decision problems could be represented in a small world decision matrix. In Savage’s words, you are in a small world if you can “look before you leap”; that is, it is feasible to enumerate all contingencies and you know what the consequences of actions are. You are in a grand world when you must “cross the bridge when you come to it”, either because you are not sure what the possible states of the world, actions and/or consequences are…
In the following three sections I elaborate on the complications mentioned above emphasizing, once again, that many real life situations are prone to such complications.
State space uncertainty
The standard view of uncertainty assumes that all possible states are given as a part of the problem definition – as in the omelet example discussed earlier. In real life, however, this is often not the case.
Bradley and Drechsler identify two distinct cases of state space uncertainty. The first one is when we are unaware that we’re missing states and/or consequences. For example, organisations that embark on a restructuring program are so focused on the cost-related consequences that they may overlook factors such as loss of morale and/or loss of talent (and the consequent loss of productivity). The second, somewhat rarer, case is when we are aware that we might be missing something but we don’t quite know what it is. All one can do here, is make appropriate contingency plans based on guesses regarding possible consequences.
Figuring out possible states and consequences is largely a matter of scenario envisioning based on knowledge and practical experience. It stands to reason that this is best done by leveraging the collective experience and wisdom of people from diverse backgrounds. This is pretty much the rationale behind collective decision-making techniques such as Dialogue Mapping.
Option uncertainty
The standard approach to tackling uncertainty assumes that the connection between actions and consequences is well defined. This is often not the case, particularly for wicked problems. For example, as I have discussed in this post, enterprise transformation programs with well-defined and articulated objectives often end up having a host of unintended consequences. At an even more basic level, in some situations it can be difficult to identify sensible options.
Option uncertainty is a fairly common feature in real-life decisions. As Bradley and Drechsler put it:
Option uncertainty is an endemic feature of decision making, for it is rarely the case that we can predict consequences of our actions in every detail (alternatively, be sure what our options are). And although in many decision situations, it won’t matter too much what the precise consequence of each action is, in some the details will matter very much.
…and unfortunately, the cases in which the details matter are precisely those problems in which they are the hardest to figure out – i.e. in wicked problems.
Preference uncertainty
An implicit assumption in the standard approach is that once states and consequences are known, people will be able to figure out their relative preferences for these unambiguously. This assumption is incorrect, as there are at least two situations in which people will not be able to determine their preferences. Firstly, there may be a lack of factual information about one or more of the states. Secondly, even when one is able to get the required facts, it is hard to figure out how we would value the consequences.
A common example of the aforementioned situation is the job switch dilemma. In many (most?) cases in which one is debating whether or not to switch jobs, one lacks enough factual information about the new job – for example, the new boss’ temperament, the work environment etc. Further, even if one is able to get the required information, it is impossible to know how it would be to actually work there. Most people would have struggled with this kind of uncertainty at some point in their lives. Bradley and Drechsler term this ethical uncertainty. I prefer the term preference uncertainty, as it has more to do with preferences than ethics.
Some general remarks
The first point to note is that the three types of uncertainty noted above map exactly on to the three shortcomings of classical decision theory discussed in the introduction. This suggests a connection between the types of uncertainty and wicked problems. Indeed, most wicked problems are exemplars of one or more of the above uncertainty types. For example, the paradigm-defining super-wicked problem of climate change displays all three types of uncertainty.
The three types of uncertainty discussed above are overlooked by the standard approach to managing uncertainty. This happens in a number of ways. Here are two common ones:
- The standard approach assumes that all uncertainties can somehow be incorporated into a single probability function describing all possible states and/or consequences. This is clearly false for state space and option uncertainty: it is impossible to define a sensible probability function when one is uncertain about the possible states and/or outcomes.
- The standard approach assumes that preferences for different consequences are known. This is clearly not true in the case of preference uncertainty…and even for state space and option uncertainty for that matter.
In their paper, Bradley and Dreschsler arrive at these three types of uncertainty from considerations different from the ones I have used above. Their approach, while more general, is considerably more involved. Nevertheless, I would recommend that readers who are interested should take a look at it because they cover a lot of things that I have glossed over or ignored altogether.
Just as an example, they show how the aforementioned uncertainties can be reduced. There is a price to be paid, however: any reduction in uncertainty results in an increase in its severity. An example might help illustrate how this comes about. Consider a situation of state space uncertainty. One can reduce- or even, remove – this by defining a catch-all state (labelled, say, “all other outcomes”). It is easy to see that although one has formally reduced state space uncertainty to zero, one has increased the severity of the uncertainty because the catch-all state is but a reflection of our ignorance and our refusal to do anything about it!
There are many more implications of the above. However, I’ll point out just one more that serves to illustrate the very practical implications of these uncertainties. In a post on the shortcomings of enterprise risk management, I pointed out that the notion of an organisation-wide risk appetite is problematic because it is impossible to capture the diversity of viewpoints through such a construct. Moreover, rule or process based approaches to risk management tend to focus only on those uncertainties that can be quantified, or conversely they assume that all uncertainties can somehow be clumped into a single probability distribution as prescribed by the standard approach to managing uncertainty. The three types of uncertainty discussed above highlight the limitations of such an approach to enterprise risk.
Conclusion
The standard approach to managing uncertainty assumes that all possible states, actions and consequences are known or can be determined. In this post I have discussed why this is not always so. In particular, it often happens that we do not know all possible outcomes (state space uncertainty), consequences (option uncertainty) and/or our preferences for consequences (preference or ethical uncertainty).
As I was reading the paper, I felt the authors were articulating issues that I had often felt uneasy about but chose to overlook (suppress?). Generalising from one’s own experience is always a fraught affair, but I reckon we tend to deny these uncertainties because they are inconvenient – that is, they are difficult if not impossible to deal with within the procrustean framework of the standard approach. What is needed as a corrective is a recognition that the pseudo-quantitative approach that is commonly used to manage uncertainty may not the panacea it is claimed to be. The first step towards doing this is to acknowledge the existence of the uncertainties that we (probably) overlook.
On the statistical downsides of blogging
Introduction
The stats on the 200+ posts I’ve written since I started blogging make it pretty clear that:
- Much of what I write does not get much attention – i.e. it is not of interest to most readers.
- An interesting post – a rare occurrence in itself – is invariably followed by a series of uninteresting ones.
In this post, I ignore the very real possibility that my work is inherently uninteresting and discuss how the above observations can be explained via concepts of probability.
Base rate of uninteresting ideas
A couple of years ago I wrote a piece entitled, Trumped by Conditionality, in which I used conditional probability to show that majority of the posts on this blog will be uninteresting despite my best efforts. My argument was based on the following observations:
- There are many more uninteresting ideas than interesting ones. In statistical terminology one would say that the base rate of uninteresting ideas is high. This implies that if I write posts without filtering out bad ideas, I will write uninteresting posts far more frequently than interesting ones.
- The base rate as described above is inapplicable in real life because I do attempt to filter out the bad ideas. However, and this is the key: my ability to distinguish between interesting and uninteresting topics is imperfect. In other words, although I can generally identify an interesting idea correctly , there is a small (but significant) chance that I will incorrectly identify an uninteresting topic as being interesting.
Now, since uninteresting ideas vastly outnumber interesting ones and my ability to filter out uninteresting ideas is imperfect, it follows that the majority of the topics I choose to write about will be uninteresting. This is essentially the first point I made in the introduction.
Regression to the mean
The observation that good (i.e. interesting) posts are generally followed by a series of not so good ones is a consequence of a statistical phenomenon known as regression to the mean. In everyday language this refers to the common observation that an extreme event is generally followed by a less extreme one. This is simply a consequence of the fact that for many commonly encountered phenomena extreme events are much less likely to occur than events that are close to the average.
In the case at hand we are concerned with the quality of writing. Although writers might improve through practice, it is pretty clear that they cannot write brilliant posts every time they put fingers to keyboard. This is particularly true of bloggers and syndicated columnists who have to produce pieces according to a timetable – regardless of practice or talent, it is impossible to produce high quality pieces on a regular basis.
It is worth noting that people often incorrectly ascribe causal explanations to phenomena that can be explained by regression to the mean. Daniel Kahneman and Amos Tversky describe the following example in their classic paper on decision-related cognitive biases:
…In a discussion of flight training, experienced instructors noted that praise for an exceptionally smooth landing is typically followed by a poorer landing on the next try, while harsh criticism after a rough landing is usually followed by an improvement on the next try. The instructors concluded that verbal rewards are detrimental to learning, while verbal punishments are beneficial, contrary to accepted psychological doctrine. This conclusion is unwarranted because of the presence of regression toward the mean. As in other cases of repeated examination, an improvement will usually follow a poor performance and a deterioration will usually follow an outstanding performance, even if the instructor does not respond to the trainee’s achievement on the first attempt…
So, although I cannot avoid the disappointment that follows the high of writing a well-received post, I can take (perhaps, false) comfort in the possibility that I’m a victim of statistics.
In closing
Finally, l would be remiss if I did not consider an explanation which, though unpleasant, may well be true: there is the distinct possibility that everything I write about is uninteresting. Needless to say, I reckon the explanations (rationalisations?) offered above are far more likely to be correct :-)
The shape of things to come: an essay on probability in project estimation
Introduction
Project estimates are generally based on assumptions about future events and their outcomes. As the future is uncertain, the concept of probability is sometimes invoked in the estimation process. There’s enough been written about how probabilities can be used in developing estimates; indeed there are a good number of articles on this blog – see this post or this one, for example. However, most of these writings focus on the practical applications of probability rather than on the concept itself – what it means and how it should be interpreted. In this article I address the latter point in a way that will (hopefully!) be of interest to those working in project management and related areas.
Uncertainty is a shape, not a number
Since the future can unfold in a number of different ways one can describe it only in terms of a range of possible outcomes. A good way to explore the implications of this statement is through a simple estimation-related example:
Assume you’ve been asked to do a particular task relating to your area of expertise. From experience you know that this task usually takes 4 days to complete. If things go right, however, it could take as little as 2 days. On the other hand, if things go wrong it could take as long as 8 days. Therefore, your range of possible finish times (outcomes) is anywhere between 2 to 8 days.
Clearly, each of these outcomes is not equally likely. The most likely outcome is that you will finish the task in 4 days. Moreover, the likelihood of finishing in less than 2 days or more than 8 days is zero. If we plot the likelihood of completion against completion time, it would look something like Figure 1.
Figure 1 begs a couple of questions:
- What are the relative likelihoods of completion for all intermediate times – i.e. those between 2 to 4 days and 4 to 8 days?
- How can one quantify the likelihood of intermediate times? In other words, how can one get a numerical value of the likelihood for all times between 2 to 8 days? Note that we know from the earlier discussion that this must be zero for any time less than 2 or greater than 8 days.
The two questions are actually related: as we shall soon see, once we know the relative likelihood of completion at all times (compared to the maximum), we can work out its numerical value.
Since we don’t know anything about intermediate times (I’m assuming there is no historical data available, and I’ll have more to say about this later…), the simplest thing to do is to assume that the likelihood increases linearly (as a straight line) from 2 to 4 days and decreases in the same way from 4 to 8 days as shown in Figure 2. This gives us the well-known triangular distribution.
Note: The term distribution is simply a fancy word for a plot of likelihood vs. time.
Of course, this isn’t the only possibility; there are an infinite number of others. Figure 3 is another (admittedly weird) example.
Further, it is quite possible that the upper limit (8 days) is not a hard one. It may be that in exceptional cases the task could take much longer (say, if you call in sick for two weeks) or even not be completed at all (say, if you leave for that mythical greener pasture). Catering for the latter possibility, the shape of the likelihood might resemble Figure 4.
From the figures above, we see that uncertainties are shapes rather than single numbers, a notion popularised by Sam Savage in his book, The Flaw of Averages. Moreover, the “shape of things to come” depends on a host of factors, some of which may not even be on the radar when a future event is being estimated.
Making likelihood precise
Thus far, I have used the word “likelihood” without bothering to define it. It’s time to make the notion more precise. I’ll begin by asking the question: what common sense properties do we expect a quantitative measure of likelihood to have?
Consider the following:
- If an event is impossible, its likelihood should be zero.
- The sum of likelihoods of all possible events should equal complete certainty. That is, it should be a constant. As this constant can be anything, let us define it to be 1.
In terms of the example above, if we denote time by and the likelihood by then:
for and
And
where
Where denotes the sum of all non-zero likelihoods – i.e. those that lie between 2 and 8 days. In simple terms this is the area enclosed by the likelihood curves and the x axis in figures 2 to 4. (Technical Note: Since is a continuous variable, this should be denoted by an integral rather than a simple sum, but this is a technicality that need not concern us here)
is , in fact, what mathematicians call probability– which explains why I have used the symbol rather than . Now that I’ve explained what it is, I’ll use the word “probability” instead of ” likelihood” in the remainder of this article.
With these assumptions in hand, we can now obtain numerical values for the probability of completion for all times between 2 and 8 days. This can be figured out by noting that the area under the probability curve (the triangle in figure 2 and the weird shape in figure 3) must equal 1. I won’t go into any further details here, but those interested in the maths for the triangular case may want to take a look at this post where the details have been worked out.
The meaning of it all
(Note: parts of this section borrow from my post on the interpretation of probability in project management)
So now we understand how uncertainty is actually a shape corresponding to a range of possible outcomes, each with their own probability of occurrence. Moreover, we also know, in principle, how the probability can be calculated for any valid value of time (between 2 and 8 days). Nevertheless, we are still left with the question as to what a numerical probability really means.
As a concrete case from the example above, what do we mean when we say that there is 100% chance (probability=1) of finishing within 8 days? Some possible interpretations of such a statement include:
- If the task is done many times over, it will always finish within 8 days. This is called the frequency interpretation of probability, and is the one most commonly described in maths and physics textbooks.
- It is believed that the task will definitely finish within 8 days. This is called the belief interpretation. Note that this interpretation hinges on subjective personal beliefs.
- Based on a comparison to similar tasks, the task will finish within 8 days. This is called the support interpretation.
Note that these interpretations are based on a paper by Glen Shafer. Other papers and textbooks frame these differently.
The first thing to note is how different these interpretations are from each other. For example, the first one offers a seemingly objective interpretation whereas the second one is unabashedly subjective.
So, which is the best – or most correct – one?
A person trained in science or mathematics might claim that the frequency interpretation wins hands down because it lays out an objective, well -defined procedure for calculating probability: simply perform the same task many times and note the completion times.
Problem is, in real life situations it is impossible to carry out exactly the same task over and over again. Sure, it may be possible to almost the same task, but even straightforward tasks such as vacuuming a room or baking a cake can hold hidden surprise (vacuum cleaners do malfunction and a friend may call when one is mixing the batter for a cake). Moreover, tasks that are complex (as is often the case in the project work) tend to be unique and can never be performed in exactly the same way twice. Consequently, the frequency interpretation is great in theory but not much use in practice.
“That’s OK,” another estimator might say,” when drawing up an estimate, I compared it to other similar tasks that I have done before.”
This is essentially the support interpretation (interpretation 3 above). However, although this seems reasonable, there is a problem: tasks that are superficially similar will differ in the details, and these small differences may turn out to be significant when one is actually carrying out the task. One never knows beforehand which variables are important. For example, my ability to finish a particular task within a stated time depends not only on my skill but also on things such as my workload, stress levels and even my state of mind. There are many external factors that one might not even recognize as being significant. This is a manifestation of the reference class problem.
So where does that leave us? Is probability just a matter of subjective belief?
No, not quite: in reality, estimators will use some or all of three interpretations to arrive at “best guess” probabilities. For example, when estimating a project task, a person will likely use one or more of the following pieces of information:
- Experience with similar tasks.
- Subjective belief regarding task complexity and potential problems. Also, their “gut feeling” of how long they think it ought to take. These factors often drive excess time or padding that people work into their estimates.
- Any relevant historical data (if available)
Clearly, depending on the situation at hand, estimators may be forced to rely on one piece of information more than others. However, when called upon to defend their estimates, estimators may use other arguments to justify their conclusions depending on who they are talking to. For example, in discussions involving managers, they may use hard data presented in a way that supports their estimates, whereas when talking to their peers they may emphasise their gut feeling based on differences between the task at hand and similar ones they have done in the past. Such contradictory representations tend to obscure the means by which the estimates were actually made.
Summing up
Estimates are invariably made in the face of uncertainty. One way to get a handle on this is by estimating the probabilities associated with possible outcomes. Probabilities can be reckoned in a number of different ways. Clearly, when using them in estimation, it is crucial to understand how probabilities have been derived and the assumptions underlying these. We have seen three ways in which probabilities are interpreted corresponding to three different ways in which they are arrived at. In reality, estimators may use a mix of the three approaches so it isn’t always clear how the numerical value should be interpreted. Nevertheless, an awareness of what probability is and its different interpretations may help managers ask the right questions to better understand the estimates made by their teams.
On the accuracy of group estimates
Introduction
The essential idea behind group estimation is that an estimate made by a group is likely to be more accurate than one made by an individual in the group. This notion is the basis for the Delphi method and its variants. In this post, I use arguments involving probabilities to gain some insight into the conditions under which group estimates are more accurate than individual ones.
An insight from conditional probability
Let’s begin with a simple group estimation scenario.
Assume we have two individuals of similar skill who have been asked to provide independent estimates of some quantity, say a project task duration. Further, let us assume that each individual has a probability of making a correct estimate.
Based on the above, the probability that they both make a correct estimate, , is:
,
This is a consequence of our assumption that the individual estimates are independent of each other.
Similarly, the probability that they both get it wrong, , is:
,
Now we can ask the following question:
What is the probability that both individuals make the correct estimate if we know that they have both made the same estimate?
This can be figured out using Bayes’ Theorem, which in the context of the question can be stated as follows:
In the above equation, is the probability that both individuals get it right given that they have made the same estimate (which is what we want to figure out). This is an example of a conditional probability – i.e. the probability that an event occurs given that another, possibly related event has already occurred. See this post for a detailed discussion of conditional probabilities.
Similarly, is the conditional probability that both estimators make the same estimate given that they are both correct. This probability is 1.
Question: Why?
Answer: If both estimators are correct then they must have made the same estimate (i.e. they must both within be an acceptable range of the right answer).
Finally, is the probability that both make the same estimate. This is simply the sum of the probabilities that both get it right and both get it wrong. Expressed in terms of this is, .
Now lets apply Bayes’ theorem to the following two cases:
- Both individuals are good estimators – i.e. they have a high probability of making a correct estimate. We’ll assume they both have a 90% chance of getting it right ().
- Both individuals are poor estimators – i.e. they have a low probability of making a correct estimate. We’ll assume they both have a 30% chance of getting it right ()
Consider the first case. The probability that both estimators get it right given that they make the same estimate is:
Thus we see that the group estimate has a significantly better chance of being right than the individual ones: a probability of 0.9878 as opposed to 0.9.
In the second case, the probability that both get it right is:
The situation is completely reversed: the group estimate has a much smaller chance of being right than an individual estimate!
In summary: estimates provided by a group consisting of individuals of similar ability working independently are more likely to be right (compared to individual estimates) if the group consists of competent estimators and more likely to be wrong (compared to individual estimates) if the group consists of poor estimators.
Assumptions and complications
I have made a number of simplifying assumptions in the above argument. I discuss these below with some commentary.
- The main assumption is that individuals work independently. This assumption is not valid for many situations. For example, project estimates are often made by a group of people working together. Although one can’t work out what will happen in such situations using the arguments of the previous section, it is reasonable to assume that given the right conditions, estimators will use their collective knowledge to work collaboratively. Other things being equal, such collaboration would lead a group of skilled estimators to reinforce each others’ estimates (which are likely to be quite similar) whereas less skilled ones may spend time arguing over their (possibly different and incorrect) guesses. Based on this, it seems reasonable to conjecture that groups consisting of good estimators will tend to make even better estimates than they would individually whereas those consisting of poor estimators have a significant chance of making worse ones.
- Another assumption is that an estimate is either good or bad. In reality there is a range that is neither good nor bad, but may be acceptable.
- Yet another assumption is that an estimator’s ability can be accurately quantified using a single numerical probability. This is fine providing the number actually represents the person’s estimation ability for the situation at hand. However, typically such probabilities are evaluated on the basis of past estimates. The problem is, every situation is unique and history may not be a good guide to the situation at hand. The best way to address this is to involve people with diverse experience in the estimation exercise. This will almost often lead to a significant spread of estimates which may then have to be refined by debate and negotiation.
Real-life estimation situations have a number of other complications. To begin with, the influence that specific individuals have on the estimation process may vary – a manager who is a poor estimator may, by virtue of his position, have a greater influence than others in a group. This will skew the group estimate by a factor that cannot be estimated. Moreover, strategic behaviour may influence estimates in a myriad other ways. Then there is the groupthink factor as well.
…and I’m sure there are many others.
Finally I should mention that group estimates can depend on the details of the estimation process. For example, research suggests that under certain conditions competition can lead to better estimates than cooperation.
Conclusion
In this post I have attempted to make some general inferences regarding the validity of group estimates based on arguments involving conditional probabilities. The arguments suggest that, all other things being equal, a collective estimate from a bunch of skilled estimators will generally be better than their individual estimates whereas an estimate from a group of less skilled estimators will tend to be worse than their individual estimates. Of course, in real life, there are a host of other factors that can come into play: power, politics and biases being just a few. Though these are often hidden, they can influence group estimates in inestimable ways.
Acknowledgement
Thanks go out to George Gkotsis and Craig Brown for their comments which inspired this post.
The drunkard’s dartboard revisited: yet another Excel-based example of Monte Carlo simulation
(Note: An Excel sheet showing sample calculations and plots discussed in this post can be downloaded here.)
Introduction
Some months ago, I wrote a post explaining the basics of Monte Carlo simulation using the example of a drunkard throwing darts at a board. In that post I assumed that the darts could land anywhere on the dartboard with equal probability. In other words, the hit locations were assumed to be uniformly distributed. In a comment on the piece, George Gkotsis challenged this assumption, arguing that that regardless of the level of inebriation of the thrower, a dart would be more likely to land near the centre of the board than away from it (providing the player is at least moderately skilled). He also suggested using the Normal Distribution to model the spread of hits, with the variance of the distribution serving as a rough measure of the inaccuracy (or drunkenness!) of the drunkard. In George’s words:
I would propose to introduce a ‘skill’ factor, which represents the circle/square ratio (maybe a normal-Gaussian distribution). Of course, this skill factor would be very low (high variance) for a drunken player, but would still take into account the fact that throwing darts into a square is not purely random.
In this post I revisit the drunkard’s dartboard, taking into account George’s suggestions.
Setting the stage
To keep things simple, I’ll make the following assumptions:
- The dartboard is a circle of radius 0.5 units centred at the origin (see Figure 1)
- The chance of a hit is greatest at the centre of the dartboard and falls off as one moves away from it.
- The distribution of hits is a function of distance from the centre but does not depend on direction. In mathematical terms, for a given distance from the centre of the dartboard, the dart can land at any angle with equal probability, being the angle between the line joining the centre of the board to the dart and the x axis. See Figure 2 for graphical representations of a hit location in terms of and . Note that that the and coordinates can be obtained using the formulas and as s shown in Figure 2.
- Hits are distributed according to the Normal distribution with maximum at the centre of the dartboard.
- The variance of the Normal distribution is a measure of inaccuracy/drunkenness of the drunkard: the more drunk the drunk, the greater the variation in his aim.
These assumptions are consistent with George’s suggestions.
The simulation
[Note to the reader: you may want to download the demo before continuing.]
The steps of a simulation run are as follows:
- Generate a number that is normally distributed with a zero mean and a specified standard deviation. This gives the distance, , of a randomly thrown dart from the centre of the board for a player with a “inaccuracy factor” represented by the standard deviation. Column A in the demo contains normally distributed random numbers with zero mean and a standard deviation of 0.2 . Note that I selected the latter number for no other reason than the results show up clearly on a fixed-axis plot shown in Figure 2.
- Generate a uniformly distributed random number lying between 0 and . This represents the angle . This is the content of column B of the demo.
- The numbers obtained from steps 1 and 2 for completely specify the location of a hit. The location’s and coordinates can be worked out using the formulas and . These are listed in columns C and D in the Excel demo.
- Re-run steps 1 through 4 as many times as needed. Note that the demo is set up for 5000 runs. You can change this manually or, better yet, automate it. The latter is left as an exercise for you.
It is instructive to visualize the resulting hits using a scatter plot. Among other things this can tell you, at a glance, if the results make sense. For example, we would expect hits to be symmetrically distributed about the origin because the drunkard’s throws are not biased in any particular direction around the centre). A non-symmetrical distribution is thus an indication that there is an error in the calculations.
Now, any finite collection of hits is unlikely to be perfectly symmetrical because of outliers. Nevertheless, the distributions should be symmetrical on average. To test this, run the demo a few times (hit F9 with the demo open). Notice how the position of outliers and the overall shape of the distribution of points changes randomly from simulation to simulation. In all cases, however, there is a clear maximum at the centre of the dartboard with the probability of a hit falling with distance from the centre.
Figure 3 shows the results of simulations for a standard deviation of 0.2. Figures 4 and 5 show the results of simulations for standard deviations of 0.1 and 0.4.
Note that the plot has fixed axes- i.e. the area depicted is the 1×1 square that encloses the dartboard, regardless of the standard deviation. Consequently, for larger standard deviations (such as 0.4) many hits will be out of range and will not show up on the plot.
Closing remarks
As I have stressed in my previous posts on Monte Carlo simulation, the usefulness of a simulation depends on the choice of an appropriate distribution. If the selected distribution does not reflect reality, neither will the simulation. This is true regardless of whether one is simulating a drunkard’s wayward aim or the duration of project task. You may have noted that the assumption of normally-distributed hits has no justification whatsoever; it is just as arbitrary as my original assumption of uniformity. In fact, the hit locations of drunken dart throws is highly unlikely to be either uniform or Normal. Nevertheless, I hope that some of my readers will find the above example to be of pedagogical value.
Acknowledgement
Thanks to George Gkotsis for his comment which got me thinking about this post.
Uncertainty about uncertainty
Introduction
More often than not, managerial decisions are made on the basis of uncertain information. To lend some rigour to the process of decision making, it is sometimes assumed that uncertainties of interest can be quantified accurately using probabilities. As it turns out, this assumption can be incorrect in many situations because the probabilities themselves can be uncertain. In this post I discuss a couple of ways in which such uncertainty about uncertainty can manifest itself.
The problem of vagueness
In a paper entitled, “Is Probability the Only Coherent Approach to Uncertainty?”, Mark Colyvan made a distinction between two types of uncertainty:
- Uncertainty about some underlying fact. For example, we might be uncertain about the cost of a project – that there will be a cost is a fact, but we are uncertain about what exactly it will be.
- Uncertainty about situations where there is no underlying fact. For example, we might be uncertain about whether customers will be satisfied with the outcome of a project. The problem here is the definition of customer satisfaction. How do we measure it? What about customers who are neither satisfied nor dissatisfied? There is no clear-cut definition of what customer satisfaction actually is.
The first type of uncertainty refers to the lack of knowledge about something that we know exists. This is sometimes referred to as epistemic uncertainty – i.e. uncertainty pertaining to knowledge. Such uncertainty arises from imprecise measurements, changes in the object of interest etc. The key point is that we know for certain that the item of interest has well-defined properties, but we don’t know what they are and hence the uncertainty. Such uncertainty can be quantified accurately using probability.
Vagueness, on the other hand, arises from an imprecise use of language. Specifically, the term refers to the use of criteria that cannot distinguish between borderline cases. Let’s clarify this using the example discussed earlier. A popular way to measure customer satisfaction is through surveys. Such surveys may be able to tell us that customer A is more satisfied than customer B. However, they cannot distinguish between borderline cases because any boundary between satisfied and not satisfied customers is arbitrary. This problem becomes apparent when considering an indifferent customer. How should such a customer be classified – satisfied or not satisfied? Further, what about customers who choose not to respond? It is therefore clear that any numerical probability computed from such data cannot be considered accurate. In other words, the probability itself is uncertain.
Ambiguity in classification
Although the distinction made by Colyvan is important, there is a deeper issue that can afflict uncertainties that appear to be quantifiable at first sight. To understand how this happens, we’ll first need to take a brief look at how probabilities are usually computed.
An operational definition of probability is that it is the ratio of the number of times the event of interest occurs to the total number of events observed. For example, if my manager notes my arrival times at work over 100 days and finds that I arrive before 8:00 am on 62 days then he could infer that the probability my arriving before 8:00 am is 0.62. Since the probability is assumed to equal the frequency of occurrence of the event of interest, this is sometimes called the frequentist interpretation of probability.
The above seems straightforward enough, so you might be asking: where’s the problem?
The problem is that events can generally be classified in several different ways and the computed probability of an event occurring can depend on the way that it is classified. This is called the reference class problem. In a paper entitled, “The Reference Class Problem is Your Problem Too”, Alan Hajek described the reference class problem as follows:
“The reference class problem arises when we want to assign a probability to a proposition (or sentence, or event) X, which may be classified in various ways, yet its probability can change depending on how it is classified.”
Consider the situation I mentioned earlier. My manager’s approach seems reasonable, but there is a problem with it: all days are not the same as far as my arrival times are concerned. For example, it is quite possible that my arrival time is affected by the weather: I may arrive later on rainy days than on sunny ones. So, to get a better estimate my manager should also factor in the weather. He would then end up with two probabilities, one for fine weather and the other for foul. However, that is not all: there are a number of other criteria that could affect my arrival times – for example, my state of health (I may call in sick and not come in to work at all), whether I worked late the previous day etc.
What seemed like a straightforward problem is no longer so because of the uncertainty regarding which reference class is the right one to use.
Before closing this section, I should mention that the reference class problem has implications for many professional disciplines. I have discussed its relevance to project management in my post entitled, “The reference class problem and its implications for project management”.
To conclude
In this post we have looked at a couple of forms of uncertainty about uncertainty that have practical implications for decision makers. In particular, we have seen that probabilities used in managerial decision making can be uncertain because of vague definitions of events and/or ambiguities in their classification. The bottom line for those who use probabilities to support decision-making is to ensure that the criteria used to determine events of interest refer to unambiguous facts that are appropriate to the situation at hand. To sum up: decisions made on the basis of probabilities are only as good as the assumptions that go into them, and the assumptions themselves may be prone to uncertainties such as the ones described in this article.