Part 2, Chapter 2: Improving the Quality of Information

Questions to ask when performing an initial inspection of imported data include:

- did column labels transfer correctly? - how many columns (variables/attributes) and rows (instances/cases) should there be and are all present? - is the data in the proper format for the system that will use the data? - are there any unintended duplicate rows? - did the values transfer correctly and are there outliers to investigate? - are there significant gaps in data and is there a pattern to the gaps?

Objectivity exists when the FP&A professional is:

- not invested in the outcome - dispassionate (unemotional) - ruled by facts and evidence - able to show integrity - consistent in responses

The purpose of identifying outliers is threefold:

- to find and correct errors - to determine whether the outlier will lead to erroneous findings and, if so, address the outlier in some way (e.g. normalize or eliminate) - to study non-error outliers for their information content

Addressing Outliers process:

- Identify outliers. - Check if outliers are errors. - Determine outlier information value. - Normalize data if possible - exclude nonrecurring from trends; otherwise retain

Process for Combining and/or Aggregating Data -

- develop proper queries for downloading data - perform initial analysis of imported data - combine and/or aggregate data

Standard Deviation variations =

65 - 95 - 99.7 rule

Failing to negotiate properly can cause a loss of goodwill or a loss of respect.

If you are too adversarial or are unwilling to budge on a number, you will damage your ability to negotiate with the same person in the future, losing goodwill. However, if you are too easily influenced and accept clearly biased information at face value, you will not be respected.

Which is an FP&A skill that is most relevant for minimizing the potential for model errors?

Intuitively noticing simple math errors such as place value errors.

In which situation is it acceptable to assume that data were downloaded correctly from an internal source?

It is never acceptable to assume that data were downloaded correctly.

An FP&A professional has previously been held accountable for taking a manager's overly optimistic scenarios for a project at face value. This time the FP&A professional takes a hard line and will not budge on a number. Which is the most likely consequence of the FP&A professional's new negotiating method?

Loss of goodwill

An organization has an unwritten expectation of constant sales growth, and it also sets bonuses for salespersons based on exceeding sales goals. When salespersons are setting their sales goals, these factors will create which of the following likely motivations?

Salespersons will set pessimistic goals just above the prior-period goals.

Data Validation

Validate data early and continuously in the process even if the task becomes onerous.

Which is the best practice for the timing of data validation?

Validate data early and continuously in the process even if the task becomes onerous.

Operational databases are the

primary data source for a model that is studying at a detailed line-item level whether all business units are providing sufficient contribution margins to the organization.

Causation bias assumes

random events are trends

If automated methods of data validation are implemented correctly, then manual methods are

still necessary.

A negative side effect of allowing a model to have poor input data is

that the same decision maker could reject a different but valid model in the future

The Z-score is

the distance of a data from the mean expressed in terms of standard deviations. It is calculated as follows x (sample) - mu (population mean) / standard deviation

Z-score is

the distance of a data value from the mean expressed in terms of standard deviations.

Z-statistic measures

the distance of a sample mean from the population mean rather than just sampling one value.

Variance is

the mean squared difference from the mean.

In a regression analysis, when an outlier is present in the data,

the regression line is a less accurate representation of the trend for most of the data.

The effect an outlier has on a regression or trend in a regression analysis or XY scatter diagram is that

the regression line is skewed toward the outlier.

The mean is

the simple average, or the sum of the values divided by the total number of values.

A standard deviation is defined as

the square root of the variance.

Optimism bias assumes

trends are random events

Garbage in, garbage out, refers to how the accuracy and reliability of the outputs are directly related to the validity, accuracy and consistency, and amount of bias in the inputs.

A model is only as good as the quality of its inputs. Sometimes this acronym is called garbage in, gospel out, to refer to the common situation in which a model's outputs often receive more credence than they merit even if the modeler provides warnings about the lack of good data inputs or the number of assumptions made.

A regression analysis plots how changing an independent variable (input) affects a dependent variable (output) to show correlation.

A regression line without the effect of an outlier might show that the data fit nicely to the regression line, but when the effect of the outlier is included, the regression line becomes skewed toward the outlier and misrepresents the trend for most of the data. Using that trend line in a model would likely produce unrealistic results.

Sometimes querying external financial information or organizational-level contribution margin income statements will be at too high a level of aggregation and it will be better to use operational data if they are available.

In this way, the results of an individual operating units might be studied individually in the model and it could reveal otherwise hidden relationships. For instance, it could show that the primary line of business actually has a poor contribution margin and is being propped up by other business units.

Which is the primary tool FP&A professionals should rely upon when checking for bias?

Reasonableness check

How can statistics be used in reference to outliers?

Statistics helps to quantify an outlier's effect on the mean or other statistical measures.

A subset of the data in a download could differ from expectations, for example, when one source system uses the metric system and another uses the English system but the unit of measure is lost when the lists are combined in a worksheet.

Such errors can be difficult to detect when examining imported data outside its original source. Therefore, checking the data in the original source(s) is an important method of data validation. Performing statistical analyses might be used to find this discrepancy in the first place, but the question indicates that the problem has already been found.

An FP&A professional provides clear indications in a model's executive summary that the model makes some big assumptions. Which is still a risk despite this warning?

The outputs may be given more credence than warranted.

What effect does an outlier have on a regression or trend line in a regression analysis or XY scatter diagram?

The regression line is skewed toward the outlier.

Z-Score =

The sample - population mean / standard deviation

FP&A professionals need to apply common sense, diligence, an intuitive knowledge of numbers, attention to detail, and business experience to detect when values seem incorrect.

They can get too comfortable relying on automation and tools to detect errors; brain power is a far more important tool. Numerical sense is an excellent example of a simple error-checking skill that requires the ability to do simple math in your head and compare the answer to what is displayed on the screen. However, FP&A professionals need to be comfortable dealing with ambiguity and should have a big-picture understanding of the organization.

One definition of conservatism is the accounting principle of immediately recognizing income-decreasing events but recognizing income increasing-events only when realized.

This type of bias requires adjusting accounting information appropriately and remaining skeptical of what organizations are reporting.

In a control chart, the upper control limit (UCL) and lower control limit (LCL) are the largest and smallest statistically expected values for the process in question.

Values between these limits fall into the range of random or natural variations. Values that fall outside these statistical limits are considered to be caused by non-random variation, or variation assignable to a specific cause. Values falling in these areas are said to be out of control, meaning they need to be investigated. In most cases these items are considered outliers.

While almost all statisticians agree that outliers are important to identify and address because they can throw off many statistical measures such as the mean, there is no well-accepted statistical definition of an outlier.

What is treated as an outlier differs depending on the context. However, statistics can be used to explain what an outlier is an to help visually identify outliers in a set of data.

Determining and interpreting outlier information value is a critical aspect of outlier analysis for FP&A professionals and is often its most important aspect.

When outliers are non-error anomalies, they are most often telling a story. Your role is to interpret this story for the benefit of decision makers. Often, the story is that outliers are just exceptionally good or exceptionally poor results that should be studied. While outliers can occur at either end of the spectrum, both types are important to study.

Failing to negotiate properly can cause

a loss of goodwill or a loss of respect. If you are too adversarial or are unwilling to budge on a number, you will damage your ability to negotiate with the same person in the future, losing goodwill. However, if you are too easily influenced and accept clearly biased information at face value, you will not be respected.

Failing to negotiate properly can cause

a loss of goodwill or respect.

An Outlier is

any data value that differs significantly from the other data values in a set of data. Outliers can be data errors, unusual events or transactions that may or may not recur in the future, or anomalies that can be studied for their information value.

When there are discrepancies between downloaded data and expectations for what should have been downloaded, first thing you need to do is

check the data at the source.

When data downloaded to a worksheet has larger differences between units than are reasonably expected but the units of measure are consistent,

check the data in the original source systems. A subset of the data in a download could differ from expectations, for example, when one source system uses the metric system and another uses the English system but the unit of measure is lost when the lists are combined in a worksheet.

FP&A professionals need to apply

common sense, diligence, an intuitive knowledge of numbers, attention to detail, and business experience to detect when values seem incorrect.

Bias in forecast can be measure after actuals are known by calculating the

cumulative forecast error = Cumulative actuals - cumulative projections

The most important part of an outlier analysis for FP&A professionals is

determining if the outlier is telling an informative story. When outliers are non-error anomalies, they are most often telling a story.

Statistics can

help quantify an outlier's effect on the mean or other statistical measures. There is no well-accepted statistical definition of an outlier.

Garbage In, Garbage Out (GIGO) refers to

how the accuracy and reliability of the outputs are directly related to the validity, accuracy and consistency, and amount of bias in the inputs.

objective information is

information that can be independently verified by all observers

Subjective information is

information that is colored by human perceptions of reality rather than being independently verifiable

When downloading data from an internal source,

it is never acceptable to assume that data were downloaded correctly.

Left skewed distribution =

mean < median < mode

Right skewed distribution =

mode < median < mean

An indispensable error checking skill related to brain power that FP&A professionals should possess is

numerical sense

Conservatism bias =

one definition of conservatism is the accounting principle of immediately recognizing income-decreasing events but recognizing income increasing-events only when realized. This type of bias requires adjusting accounting information appropriately and remaining skeptical of what organizations are reporting.

In a control chart, non-random or assignable variations are examples of

outliers. In a control chart, the upper control limit (UCL) and lower control limit (LCL) are the largest and smallest statistically expected values for the process in question. Values between these limits fall into the range of random or natural variations. Values that fall outside of these statistical limits are considered to be caused by non-random variation, or variation assignable to a specific cause. Values falling in these areas are said to be out of control, meaning they need to be investigated. -- outliers

The primary method of checking for bias is to

perform a reasonableness check when persons provide data or assumptions or when decision makers request changes. This combination of common sense and expertise can be performed as the information is submitted, while quantitative measurements of forecast error cannot be conducted until after actual results are available. Confirmation and anchoring are types of bias.

Normalizing data refers to

recasting financial information to include various adjustments, credits/debits or error corrections in the periods in which they would have applied if they had been recorded correctly in the first place (without the need for estimation or later adjustment).

Z-score =

sample - population mean / standard deviation

Interquartile range is a

statistical method used to exclude outliers. It is the range of the center 50% of the values in a set of data.

The median is

the middle value in a sequential list of values (if there are five values, the median is the third value [two on each side])

The mode is

the most frequently occurring value; in a frequency distribution, it will always be the highest point in the curve (unless there are two or more modes).

While almost all statisticians agree that outliers are important to identify and address because they can throw off many statistical measures such as the mean,

there is no well-accepted statistical definition of an outlier.

Objectivity is an

unbiased mental attitude that requires FP&A professionals to exercise their expertise and best judgment when determining what should reasonably be expected and not subordinate that judgment to the influence of others if doing so would compromise objectivity.

What is the next step to take if a subset of data downloaded to a worksheet has larger differences between units than are reasonably expected but the units of measure are consistent in the worksheet?

Check the data in the original source systems.

Which type of bias that might exist in an external organization's financial statements requires the statements to be adjusted to reflect the organization's actual economic income?

Conservatism

What is often the most important part of an outlier analysis for FP&A professionals?

Determining if the outlier is telling an informative story.

When checking for bias,

FP&A professional should use a reasonableness check.

Which should be the primary data source for a model that is studying at a detailed line-item level whether all business units are providing sufficient contribution margins to the organization?

Operational databases.

In a control chart, non-random or assignable variations are usually examples of what?

Outliers

PDCA stands for

Plan, Do, Check, Act