Context matters as much as content. After all, what does the content mean in isolation? Frequently, nothing. The context of a message can make it a lie just as much as inaccurate content.
I like maps. Some maps are neutral, such as maps that show where in the US people say pop, soda, soda pop, and coke.
What about a map of hateful tweets? Let's set aside the issues of defining hateful as well as false positives and negatives. The map itself, even with fully accurate information, can be misleading. How? Population, density, and access. A hateful tweet in an area of low population and/or low connectivity is a larger proportion of the overall population, and may therefore be considered more representative, than a single hateful tweet in an area of high population. Frequency maps tend to look like population maps.
This can be corrected a few ways. One would be to give a ratio, such as hateful to neutral or positive tweets. Another would be to divide by the population of an area, so that it becomes tweets per person, though that will give lower weight to tweets in areas of lowered connectivity.
I also like graphs. They can say so much. They can give levels, trends, and context to what is happening. But that Y-axis is a trouble-maker. Maybe you're dealing with numbers in the millions, where the differences are significant, but if you have to stretch all the way down to zero, they'll get shrunk to visible insignificance. Rescaling can help, but if the differences are small relative to the overall size, then they'll still vanish. Starting somewhere other than zero, or using a broken axis can fix this. Yet it creates its own problems. Anything can appear big, or small, if you pick the right axis. Going from $1,000,000 to $1,000,001 is surely insignificant, but if you start the axis at $999,999 then it will look huge; visually the one dollar change looks like it is double.
Even the numbers themselves can be tricky. Keeping the million to million and one dollar scenario, what can we call that? Maybe last year the relevant number was nine-hundred ninety-nine dollars and it had been so for decades before. In that case, hasn't the gap between last year's number and the historical trend been doubled? Alternatively, it's a 000.1% increase from the year before (I hope I counted my zeros correctly). And look, over there is two million, so we're at half that! Whatever that is.
These are not all cut and dry. Maybe the change does matter despite the seemingly small magnitude and needs to be magnified. Sometimes space is an issue. Yet the question should always remain: Does the narrative exist outside of this graphic or is it created entirely from the display?
I like maps. Some maps are neutral, such as maps that show where in the US people say pop, soda, soda pop, and coke.
What about a map of hateful tweets? Let's set aside the issues of defining hateful as well as false positives and negatives. The map itself, even with fully accurate information, can be misleading. How? Population, density, and access. A hateful tweet in an area of low population and/or low connectivity is a larger proportion of the overall population, and may therefore be considered more representative, than a single hateful tweet in an area of high population. Frequency maps tend to look like population maps.
This can be corrected a few ways. One would be to give a ratio, such as hateful to neutral or positive tweets. Another would be to divide by the population of an area, so that it becomes tweets per person, though that will give lower weight to tweets in areas of lowered connectivity.
I also like graphs. They can say so much. They can give levels, trends, and context to what is happening. But that Y-axis is a trouble-maker. Maybe you're dealing with numbers in the millions, where the differences are significant, but if you have to stretch all the way down to zero, they'll get shrunk to visible insignificance. Rescaling can help, but if the differences are small relative to the overall size, then they'll still vanish. Starting somewhere other than zero, or using a broken axis can fix this. Yet it creates its own problems. Anything can appear big, or small, if you pick the right axis. Going from $1,000,000 to $1,000,001 is surely insignificant, but if you start the axis at $999,999 then it will look huge; visually the one dollar change looks like it is double.
Even the numbers themselves can be tricky. Keeping the million to million and one dollar scenario, what can we call that? Maybe last year the relevant number was nine-hundred ninety-nine dollars and it had been so for decades before. In that case, hasn't the gap between last year's number and the historical trend been doubled? Alternatively, it's a 000.1% increase from the year before (I hope I counted my zeros correctly). And look, over there is two million, so we're at half that! Whatever that is.
These are not all cut and dry. Maybe the change does matter despite the seemingly small magnitude and needs to be magnified. Sometimes space is an issue. Yet the question should always remain: Does the narrative exist outside of this graphic or is it created entirely from the display?
No comments:
Post a Comment