Discussion


Note: Images may be inserted into your messages by uploading a file attachment (see "Manage Attachments"). Even though it doesn't appear when previewed, the image will appear at the end of your message once it is posted.
Register Latest Topics
 
 
 


Reply
  Author   Comment   Page 2 of 3      Prev   1   2   3   Next
GiovanniMilan

Registered:
Posts: 4
Reply with quote  #16 
To return to Stephen's article, please consider the graph I proposed in the post #8.
It is a scatter plot with the same axes, in which the vertical distance of each point from the y=x line shows the magnitude of change. It is similar to a range bar graph (see on page 3 of the article) just rotated 90° counterclockwise. But, in a range bar graph you waste one of the axes to show the ranks, instead of the distances among the items.
I think that mine is better than range bar graph and than a slope graph, because it shows both pre- and post-ranks and both pre- and post-distances (as slope graphs do), and furthermore it overcomes the following drawbacks of slope graphs:
1) In a slope graph, variations are coded by ...slopes, that is an element that you can't assess accurately. On the contrary, in my graph they are coded by segments.
2) In a slope graph, the aspect ratio is quite arbitrary: you can stress the magnitude of all changes by moving the lines closer, and vice versa. On the contrary, you can't cheat using my graph because it is a perfect square.
3) In a slope graph, slanted lines are difficult to follow when they intersect. On the contrary, there are no intersecting lines in my graph.
4) In a slope graph you show the same labels twice in redundancy. On the contrary, in my graph they are shown just once.

My two-penn'orth.
AndrejLapajne

Registered:
Posts: 4
Reply with quote  #17 

Hi,

Here is another solution, displaying actual values from 2013 and the change from previous year in a combined graph (aka “integrated deviation”):

Zebra-BI_Variance_Analysis_light.png 

Of course, other shapes and colors can be applied (as long as they obey the rules of clarity). Usually a normal bar shape would be used, arrows are just an example to demonstrate a different option. Here a slightly more neutral design and this time labelled with relative deviations:

Zebra-BI_Variance_Analysis_relative.png      

I intentionally displayed the labels. Labels of actual values are on the left, vertically aligned in a column but still next to bars. Deviation labels are on the right of the deviation shapes, so that they can be read without unnecessary eye saccades or even additional lookups.

This is a combination of a range bar graph and a basic bar chart. The advantage it over the range graph, depicted in the newsletter, is that the reader can:

  • Immediately understand the rank
  • Easily compare magnitude of values (“importance” of products)
  • Easily compare absolute deviations (changes) between different products, but also
  • Understand the relative variance (growth in %) by observing the relation between the deviation and the absolute value (revenue) for each product
It is efficient in terms of consumption of space and it also allows efficient labelling of both values and deviations (especially compared to the “slope” graph) and possible further enhancements (if needed), such as single category highlights, difference highlights, display of averages (e.g. by a vertical dotted line), benchmarks, etc).

I have been using this chart for over 8 years now and it was received extremely well by our clients. All credits for it go to dr.Rolf Hichert, who is (at least to my knowledge) the inventor of this chart.

Let me add a few observations to the thoroughly presented topic in the newsletter:

Line graph (“slope”)

Spotting changes in rank is the greatest strength of this chart, clearly visible by lines crossing. However, in the presented example, comparing values at the same point in time is not easy because the axis is cut. For example, the perception of Regular Espresso and Amaretto in 2012 is that the revenues of Amaretto are approximately 50% higher than Regular Espresso. This is of course not true because the axis begins at 10,000 instead of 0, resulting in lie factor of 2,53 in this case (Espresso represents 85% of Amaretto sales, while it is perceived as 33%). This could be solved by either

·        deleting the X axis and moving year labels to the top, thus focusing users to observing ranks and trends only or

·        by not cutting the axis which would also allow the users to compare values between products in each year without misinterpretations.

While it is true that coloring each line would spoil the legibility, coloring classes usually helps. For example, in this case I would propose to color code teas and coffees. By doing that it becomes apparent that all coffee products are growing, and that all three products with decrease in sales are tea (Lemon, Darjeeling, Mint):

Slope graph_andrej-lapajne.png 

In this redesign, I have also deleted the value axis and introduced number labels next to category labels. This is according to Tufte’s proposition for this chart from 1983. Why would you introduce value axis? It is not helping. It forces observers’ attention to shift left and right between the data points and the axis line, markers and axis labels, trying to guess what the values might be. This is unnecessary effort, which can in best scenario result only in an approximate guess of what the revenues might be.

Logarithmic scale

Logarithmic scale allows comparing rates of change, but this practice is dangerous, especially if the non-linear scale is not clearly visually marked (as in this case). 

Non-linear representations can be very useful in some cases and for a particular audience. In scientific visualization, users are accustomed to log scales, while majority of “regular” users and business users (managers) could easily misinterpret this data as “linear” sales. Especially because the axis looks exactly the same as the linear one. The only difference is in the content of class labels. Until you do not read them, you don’t know that this is in fact a log scale. At least the tickmarks in a non-linear pattern (e.g. for every 10.000 USD) should be added, but for business data like Revenues it is better to avoid log scales completely.

Range bar graph

I think the short lines improve the understanding. The problem of this example is that the axis should definitely start with 0 (not be cut). I believe this is a must in all bar charts. Perhaps some users will only observe the magnitudes of changes, but others will also perceive magnitudes of values, as it is suggested in the article itself.

E.g. can you tell what is the value (revenue) of Reg. Espresso compared to Amaretto? Although this time the X axis is not displayed, the perception of the baseline is somewhere between the end of product labels and beginning of gridlines. This is not correct, resulting in a lie factor. The relations between values are visually “magnified”.

For example, it seems that Caffe Latte grows at more than 250% rate, while the true growth rate is only 112% (clearly visible on my second example above). If the axis would start with 0, then also the perception of the change compared to the absolute value (revenue) would be correct. This would enable correct interpretations of relative changes as well. 

1st Combination of graphs

A combination of charts is even more meaningful if it allows easy comparisons of the same types of values across charts. In this example, the same KPI (Revenues in USD) is displayed with two different shapes (lines on the left, bars on the right) and orientations, which requires longer cognition time and prevents direct comparisons. Scaling the value across across both charts would also help.

2nd Combination of graphs

This time the situation is reverse. Two different types of values are presented (revenues in thousands of USD, %), but the same visual variables are used: same shape, same width, same (?) color. It two different shapes would be used, the user would immediately understand that he is looking at two different types of values. In this example, this realisation only occurs after reading the chart titles and axis labels. It would be better to code it in graphics because visual perception is immediate or at least much faster.

Using "narrow” shapes for percentages is the most intuitive way, because a percentage does not have a "unit" like USD, Pcs, Employees, etc. Technically the unit is %, but this is just a representation of a number without a unit. If you think about it, the growth is calculated as USD in this year, divided by USD in previous year. When you perform the division, the unit (USD) goes away, and the result is a “smaller” number (index, percentage). In visual representation (at least of business data), the division of two values with the same unit should result in shapes where the width is not perceived (only the length is). I call this the “division of shapes” and in my opinion, the optimal resulting shape is a dot plot, preferably with a drop line. Dr. Hichert calls this a pin, while the trendy name for it is “lollipop” ;)

Number Labels

The amount of number labels displayed depends on the purpose/intention of visualization. For example, is it just to illustrate a simple message in a live presentation that has to be clear in a few seconds to a general audience? Or is it a report that a sales manager will study for several minutes to fully understand sales results? Many factors are involved: user’s previous knowledge, time available for perception and cognition, context of communication, exploration vs. communication, etc.

All examples discussed in the article refer to business data (revenues, revenue growth), which significantly increases the chance that numbers should be displayed in expected precision (thousands of USD, percentages to one decimal). In fact I would argue that wrong labelling is one of the key reasons for poor acceptance of visualization by managers.

Displaying the right amount of number labels in the right position in is one of the hardest problems in visualization and reporting. It is relatively easy in vertically oriented charts, but gets much harder in horizontal charts, stacked charts or even scatter plots and bubble charts. Unfortunately just declaring that numbers must not be displayed and that value axes should be present will not solve the problem.

IBCS standard proposes many solutions for labelling and other issues in business communication (visualization and reporting). Although it might still contain some suggestions that are not optimal (in particular the overlapped charts and perhaps the excessive use of black color and markers) it is already a comprehensive and impressive body of knowledge that has been developed for many years and proved in practice by numerous companies. It follows well established theories of visualization and graphic languages by E. Tufte, J. Bertin and other authors. But it has to be clear that IBCS is not trying to address only the issues of clarity and legibility, but rather a much bigger problem of developing a sign system for a specific purpose (the letter B in IBCS stands for Business). Many companies around the world would benefit from such a standard in a similar way that drivers benefit from a standard traffic sign system. The IBCS proposal is currently in consultation draft stage so everybody is invited to participate with suggestions and proposals.


Best regards!

Andrej



__________________
Andrej Lapajne
danz

Registered:
Posts: 182
Reply with quote  #18 
@Giovanni

Basically you used a correlation graph (Revenue 2013 vs Revenue 2012) and you added for delta a colored line which connects (x=2012,y=2013) dot with (x=2012, y=2012) location (which is on the diagonal line if both scales are the same). While it brings a slight improvement over a simple correlation chart by showing more clear the variation with a colored line it is a difficult task to compare those lines without having the same baseline and not being equally spaced. More than that, your design would become very difficult to read if similar values are encountered in the same graph. 


@Andrej,

A range chart (also known as floating bar chart) does not need to be zero based scaled. If variations would be in the limit of, let's say, 5-10%, you would waste plenty of space to fit a range graph within available space. I personally find floating bar chart more useful with time scale (known as gantt chart) where the origin of time is usually the start date of a project or any other arbitrary date before that, making the sense of the zero (origin) just an conventional reference.

A second, separate deviation chart has several reasons:
1. It uses the same baseline (far superior comparison).
2. It uses its own scale based on absolute change rather than absolute value (improves readability).
3. It removes the clutter from the combined graph letting both of them easier to scan. 

When you combine two graphs you need to bring some extra elements to make the two information more distinct. This way you scan the same graphs looking for: 1st Length of the 2013 revenue (gray bars) and 2nd: change (2013-2012) in absolute value and percentage.
For this you came up with new graphical elements: Thin line (2012 revenue, as origin of change), direction of change (arrow shape and color - some redundancy here), different label format (italic vs regular) for percentage variation. Your graph is just a variation of Dr Rolf Hichert approach, both of them look nice, they do provide useful information, but I found them more difficult to read than two or three distinct graphs next to each other with their own scales and labels. Using different encoding techniques (eye pleasant or not) for different information inside same graph it will always bring some complexity to the reader to quickly decode them.

I agree that logarithmic scales have to be used with care, I found only isolated situations where logarithmic scales were appropriate and those were only related to dot or correlation charts.


@Stephen

The slope graph looks for me the least useful from the perspective of different datasets I had to deal with during time. Except some rare situations, several values can be similar, making the automatic labeling impossible. It is possible to have specific situations where values are distinct enough to choose for such of representation, but I do prefer the other methods you showed in your article.

I also want to mention that special attention is required when dealing with extreme or special values which would ruin the readability of graphs in any situation. When I deal with similar analysis for Customers or Product Category and few of them are newcomers at the end of the previous period or more recent, the percent change goes to huge values (or even infinite). In such of case I usually limit the %change scale from -100% to 100% (or 200% or more if it has sense) and mark the outliers outside the scale. For such of situations I prefer to use a dot graph instead of bar chart and a special mark or color for outliers.

Dan
AndrejLapajne

Registered:
Posts: 4
Reply with quote  #19 
Hi Dan,

I completely agree with your point on gantt chart. There is no problem with lie factor there, because usually the baseline is at the start of the first task. Even if it is not, the perception is still correct, because all values (start, end and duration of tasks) are displayed according to the time axis, nothing is cropped.

In Stephen's example I was referring to the text below the graph, that it supports "comparing the magnitues of values". This statement is only true if the scale starts with 0. If the axis is cropped, then the function of this graph becomes limited to comparing the magnitudes of change. If somebody visually compares the magnitudes of values, he will make a mistake. Also the perception of the relative importance of the change is spoiled. The mistake is bigger for small values and decreases with larger values. So I think it would be much better not to cut the axis in this case. Saving space is very important, but in my opinion, the correctness of interpretation has priority over the usage of space.

I also agree that separate charts are easier to comprehend. They consume more space but you gain advantages that you are describing. Both solutions are valid and the decision, which one to use should be based on target audience and other criteria, such as how much space is available, etc. Also the decision regarding how much complexity is required should be very conscious. In case of business data, some more complexity is usually OK, as long as the shape and color coding is strictly consistent. Especially if users will receive reports over a longer period of time. That's why standards are so important.

I completely agree that the slope graph is the least useful of all the options discussed.

You have a very good point by mentioning the problem with outliers in the case of relative deviations. IBCS proposes exactly the same solution: dot graph and marking outliers outside the scale with a triangle shape. The triangle is directed left or right, depending on the positive/negative deviation.

Regards!
Andrej


__________________
Andrej Lapajne
danz

Registered:
Posts: 182
Reply with quote  #20 
Andrej,

The magnitude of values are given by the start-end of the floating bars. In gantt the bars encode start, length and end of a task, and we do not have any issue in confusing any of them. If you would consider instead of time scale another quantitative scale should make no difference. If you would use a dot chart with two diferent marks for start and end would you need a zero based scale?

Long time before IBCS appeared, Stephen advocate the need of zero based scales for bar charts. This does not apply to dot charts, correlation charts, candle charts, box plot charts, line charts or floating bars (range charts), but it applies to bar charts and area charts for obvious reasons.

As I said before if you would have a very small variation (less than 5% for all categories) what is the catch of using this kind of chart? Close to none, because with an absolute scale the range bars can be very small. If you would use the min/max values to fit all in, the bars (position and length) would provide the right information to decode start value, end value and variation. Not perfect for variation (not sharing same baseline), with extra graphical encoding (for sign), but decent enough.

Dan

(It seems that templates 7 and 8 are missing at provided link)
mcarper

Registered:
Posts: 5
Reply with quote  #21 
I'm leaning towards Andrej's criticism of the range bar graphs. Because when I read the axis title, "Change in product revenue," and then see the axis tick marks, which say 10000, 20000...I assume that those numbers are the change. Because bar charts use the end point of the bar on the axis to show their measure. But here, change isn't denoted by the end point of the bar, but the difference between the ending point and the starting point. 

In other words, the title makes me thing this will be a deviation chart, because in that case, the axis is the change. 

Stephen, I'm wondering if there's something you can do to the axis to make it clear that it shows values, not change. 

__________________
-Michael
AndrejLapajne

Registered:
Posts: 4
Reply with quote  #22 
Hi,

@danz:
Even though bars are more explicit, the shape itself (dot vs. bar) is not crucial for the perception of absolute values. The distance from the axis to the position of the value (end of the bar or position of the dot) is. The moment you cut the axis, the perception of the values does not match the actual numbers and the chart does not allow proper value comparisons any more. Not only that, it implies wrong values. The only function of such a chart remains to compare changes. In this case you should rather change the chart to the deviation ("plus-minus") chart (calculate and display the deviations only) instead of cutting the axis.

If you have a very small variation, then the message of such a chart is exactly this: "the variations are very small". This can be a perfectly valid message. Many times, people feel they have to display large variances (so that "the users will see them"), even if the variance is close to none... This always results in wrong (exaggerated) graphics. Such a visualization completely changes the story that the numbers are saying. That is why it is important to display not only one set of numbers, but also their context. High information density is important and the decision what to display should not be based solely on the data, but rather on the amount of information that the intended audience (users) is able to perceive and process.

So regarding the zero based scale: I think charts should always have a zero based scale, except sometimes in line charts if it is beneficial and if:
  1. the only intended message of the line chart is to observe and compare trends
  2. the axis is completely deleted (the line of the axis and preferably also the axis labels)
@mcarper:
The problem of the title is, that is refers to only one invariant of the range chart (changes), while in fact the chart has two invariants (values, changes). So this title is not complete. The title should always state all the invariants of the chart (in some cases also components - if they are not labelled beside the chart elements). In my redesign this is "Revenues" (above the chart) and also "ACT" (or "2013") and "∆PY" (or "Growth from PY") above the respective values.

Instead of doing something to the axis, displaying the numbers (label the changes beside the bars) would be much more beneficial. At least in my humble opinion... ;)

Best regards to the community!
Andrej



__________________
Andrej Lapajne
danz

Registered:
Posts: 182
Reply with quote  #23 
Andrej

I agree that using a deviation chart is more beneficial than a range graph when we study the variation only, but I also consider that a range graph has extra information providing start-end values.

I share with you the concern of wrong designed bar charts with an origin different than zero. But I think you interpret too strict the zero based charts rule. More than that, you deliberately ignore the presence of the scale. A range bar chart intention is not to compare the start, end or length against zero axis, but to compare start, end and length between items using the associated scale. I see no vertical line in Stephen graph to be misinterpreted as origin, but I see very clear the top scale which I use to decode the values. 

I agree that a variation should be represented at the same scale with absolute values, but only if they need to share the same chart and this can happen only when we need to save some space. I sustain the logic of keeping the same scale across trellis charts, wrapped bar chart or other multi chart combinations across different pages of the same document when they share the same measure in different contexts. But in rest any audience should be trained to interpret a chart as it was designed, with it's own scale, not related to the other charts scales within same screen, page. It is no exaggeration involved in using a proper (fit) scale for small or large values. If you would design a sales dashboard where you need to see four graphs for clients, products, product categories, sales agents, would you use a common scale for all of them?
AndrejLapajne

Registered:
Posts: 4
Reply with quote  #24 
Hi danz,

no problem, I think you are missing a point with the scale on range chart, but we do not have to agree on everything ;) Let's leave the space for other people's comments.

To answer your question on scaling charts in a sales dashboard: it depends on the structure and composition of the dashboard, but yes, in most cases I would. Sometimes this is not meaningful, but at least I always think about it. E.g. I would scale products and clients or product categories to sales agents (different dimensions, similar level of detail) if I would estimate that value ranges are similar or that the reader will likely observe and compare them. Especially if those charts explain the breakdown of a larger number (e.g. sales totals or total budget variance displayed at the top of the dashboard). Product categories to products (same dimension, different level) is a different story:
  1. sometimes (less often...) I would scale them and deliberately use more space for product categories
  2. if I decide not to scale them, then I try to use scale helpers (so that user understands the magnitudes in the parent-child relationship)
A.

__________________
Andrej Lapajne
jlbriggs

Registered:
Posts: 191
Reply with quote  #25 
Quote:
Originally Posted by AndrejLapajne

So regarding the zero based scale: I think charts should always have a zero based scale, except sometimes in line charts if it is beneficial and if:
  1. the only intended message of the line chart is to observe and compare trends
  2. the axis is completely deleted (the line of the axis and preferably also the axis labels)


Do I understand correctly that if you had a line chart with these values:

chart(1).png 

You would either force a meaningless zero-min like this:
chart(2).png 
Or remove the y axis so that the user cannot determine the range of values like this:
chart(4).png   
?

Or have I misread your stance?

mcarper

Registered:
Posts: 5
Reply with quote  #26 

Quote:
Originally Posted by AndrejLapajne

@mcarper:
The problem of the title is, that is refers to only one invariant of the range chart (changes), while in fact the chart has two invariants (values, changes). So this title is not complete. The title should always state all the invariants of the chart (in some cases also components - if they are not labelled beside the chart elements). In my redesign this is "Revenues" (above the chart) and also "ACT" (or "2013") and "∆PY" (or "Growth from PY") above the respective values.

Instead of doing something to the axis, displaying the numbers (label the changes beside the bars) would be much more beneficial. At least in my humble opinion... ;)

Best regards to the community!
Andrej




I'm still not sure adding both variants to the title would clarify Stephen's Range Bar graph. In fact, I think it would be best to simply label it "Product Revenue, 2012 & 2013." Because otherwise, if you say "change," the user associates the "change" value with value on the axis. Even if you said "product revenue and revenue change," it's not readily apparently which value the axis refers to. Decaf Espresso is at 40,000--but 40,000 what?

I do think I prefer your rendering, which uses full bar graphs, as opposed to this one. Because even if it was titled as I spelled out, the concept of bars floating on an axis, instead of starting at 0, contradicts how we read bar graphs. Because we can't rely on the visible length of the bar to gauge its value. Perhaps there's something wrong with me, but I can't get this graph to "click."


__________________
-Michael
jlbriggs

Registered:
Posts: 191
Reply with quote  #27 
Quote:
Originally Posted by mcarper

...even if it was titled as I spelled out, the concept of bars floating on an axis, instead of starting at 0, contradicts how we read bar graphs. Because we can't rely on the visible length of the bar to gauge its value. Perhaps there's something wrong with me, but I can't get this graph to "click."



Floating bars do not contradict how we read a standard bar chart in any way at all.

The reason that a bar chart must have a zero baseline is that we need to see the full length of the bar in order to understand what it represents. In a standard bar chart, we use a common baseline because that makes it easiest to compare the individual values, and that baseline must be zero in order to display the full length of the bar.

With a floating bar chart, you ARE seeing the full length of each bar. Each bar represents a set of values from low to high - the full extent of the values encoded in the bar.

We do not use a common baseline, because each floating bar does not encode an individual value - it encodes a range or a distribution.



mcarper

Registered:
Posts: 5
Reply with quote  #28 
Quote:
Originally Posted by jlbriggs
Quote:
Originally Posted by mcarper

...even if it was titled as I spelled out, the concept of bars floating on an axis, instead of starting at 0, contradicts how we read bar graphs. Because we can't rely on the visible length of the bar to gauge its value. Perhaps there's something wrong with me, but I can't get this graph to "click."



Floating bars do not contradict how we read a standard bar chart in any way at all.

The reason that a bar chart must have a zero baseline is that we need to see the full length of the bar in order to understand what it represents. In a standard bar chart, we use a common baseline because that makes it easiest to compare the individual values, and that baseline must be zero in order to display the full length of the bar.

With a floating bar chart, you ARE seeing the full length of each bar. Each bar represents a set of values from low to high - the full extent of the values encoded in the bar.

We do not use a common baseline, because each floating bar does not encode an individual value - it encodes a range or a distribution.





I suppose, I'm just having trouble seeing a single bar as representing the magnitude difference between two periods, instead of the magnitude of a single period, which what it's usually used for.

__________________
-Michael
danz

Registered:
Posts: 182
Reply with quote  #29 
Michael,

I mentioned above a gantt chart as an example of graph with no origin. A gantt chart is a floating bar chart where horizontal scale is time and is used to encode the difference in time between two events, as well as the start and the end of a task. Usually a gantt chart connects start end with other tasks via other lines to suggest logic dependencies. A time scale is basically a quantitative scale. Other database software implements "interval" as independent column type just for the convenience of time units calculations. But all the math operations with those are the same as any numbers. 

If that example is not enough, take just a moment and look at the OHLC graphs (open-high-low-close) for stock variation. If you ignore high and low values, only open-close values are left. Depends on implementation, the rectangle is filled or not or is colored different if variation is positive or negative. Essentially is a floating bar. And it has no sense at all to be a zero based chart. 

Zero based charts rule applies only to regular bars which encode values by length or area charts for similar reasons.

A line chart, dot chart, floating bar chart, box-plot, ohlc graphs, correlation graphs, encode different values using the coordinate of the scale, not using the distance till one axis. If the representation has some lengths as floating bars do, the length does not encode a value related to the origin of the scale, but related to the distance between ticks of that scale. Obviously the decoding process for the length of the bars takes longer than other charts, this is why you see lower scores provided by Stephen in his article. 


Dan


mcarper

Registered:
Posts: 5
Reply with quote  #30 
Quote:
Originally Posted by danz
Michael,

I mentioned above a gantt chart as an example of graph with no origin. A gantt chart is a floating bar chart where horizontal scale is time and is used to encode the difference in time between two events, as well as the start and the end of a task. Usually a gantt chart connects start end with other tasks via other lines to suggest logic dependencies. A time scale is basically a quantitative scale. Other database software implements "interval" as independent column type just for the convenience of time units calculations. But all the math operations with those are the same as any numbers. 

If that example is not enough, take just a moment and look at the OHLC graphs (open-high-low-close) for stock variation. If you ignore high and low values, only open-close values are left. Depends on implementation, the rectangle is filled or not or is colored different if variation is positive or negative. Essentially is a floating bar. And it has no sense at all to be a zero based chart. 

Zero based charts rule applies only to regular bars which encode values by length or area charts for similar reasons.

A line chart, dot chart, floating bar chart, box-plot, ohlc graphs, correlation graphs, encode different values using the coordinate of the scale, not using the distance till one axis. If the representation has some lengths as floating bars do, the length does not encode a value related to the origin of the scale, but related to the distance between ticks of that scale. Obviously the decoding process for the length of the bars takes longer than other charts, this is why you see lower scores provided by Stephen in his article. 


Dan




Thank you Dan. I would just like to clarify what exactly I find non-intuitive about the range bars. It's not merely that it uses a non-zero axis, but that it uses blank space, where series should have  relevant values. Neither gantt bars nor OHLC charts suffer from drawback.

[1] 

So for "Training," there are no associated values for 08/07. Training isn't relevant until (ie it starts at) 23/08-ish, so it's easy to comprehend why the bar starts there.

[ThickBars] 

Likewise, for the final minute of the day, the stock price never occupied the 65.60 price. It has no relevant values below 66.10, so it's easy to comprehend why it's floating.

However, with Stephen's example:

range bar graph few.PNG 

Colombian coffee does, in fact, have values all the way up through 65,000. And it's not made obvious by the chart why the bar only starts at that point. That, or I'm simply so used to seeing bar charts used a certain way that I can't easily switch to different use.


__________________
-Michael
Previous Topic | Next Topic
Print
Reply

Quick Navigation: