From Newsgroup: comp.sys.mac.advocacy

On 11/13/2025 10:52 AM, -hh wrote:

On 11/10/25 19:04, Tom Elam wrote:

On 11/8/2025 9:26 PM, -hh wrote:

No, you'd provided a simple plot that was allegedly reliant on just
one independent variable of 'temperature', and because you had only
one independent variable, that made it univariate instead of
multivariate. For you to now claim herein to have had more variables
all along is dishonest.

Yes, the chart is just temp and kWh. It is not the product of multiple
regression with many more variables. All the chart shows is that this
is more like a quadratic than linear relationship.

Because of the high level of noise, it wasn't an immediately obvious conclusion:  it required having an understanding of the nature of the question to recognize how on first principles it shouldn't be linear.

It was the time-saving starting point before the multiple regression
was performed immediately after and to determine a starting point for
that first run. That was all done about 5 years ago.

Your data reaches back to 2010: what utility companies provide a 10+
year archive on historical billing information?

My archive, not AES. I had paper copies of bills back to 2010 in a file drawer. AES has a 3 year history on their site too.

Sounds far more likely to me that you started this ~15 years ago and
have been merely maintaining it since then.

Right, but updating monthly and looking for better results as data
accumulated and changes made in the house.

One of the basic items covered early in the first university
statistics classes is to graph your data BEFORE you do anything else.
You never took that first class did you?

You've not taken any classes since 1973, have you?  /s

No, but I taught statistics. In recent years I have updated via
internet classes.

By the time that I started taking statistics classes, there were
handheld calculators (& later, computers), so curriculum methods
adapted.  Plus there were post-graduate classes which helped with the
real world problems.

So you did take stats classes. Good to know.

But more important AFAIC was the applied work.  Encountering novel challenges meant going to a sister organization's statistics office:
they had several statisticians that I was able to consult and learn from...and vice versa: when I'd come in with a new challenge, it wasn't uncommon to have a "huddle" of 3-4 specialists in different fields of statistics for which to figure out which lane of statistics was most promising to answering the question.

We did the same thing at Eli Lilly, with multiple stats teams working
with clinical trials and quality control data.

Here is an excellent article on why we make simple graphs like this.

https://nastengraph.medium.com/why-you-should-always-visualize-your-
data-first-18b8f432bc14Unfortunately, visualization isn't an "always"
as the above claims.

One uses the tool when appropriate, such as when you don't have a clear understanding of the nature of the response, for that's when it can be helpful.  But when you already know the form, its a waste of one's time.

In this case it was not. I used Excel to verify the first regression
form is quadratic via the graph's right-click "add trendline" command.
From there added the other variables via the Data Analysis add-in.

I rented the Analyze-It Excel add-in for client work and was able to go
well beyond Excel's 16 independent variable limit and take advantage of
many other features not present in Excel. Turned out that was not needed
for this fun little exercise so I dropped the subscription.

https://analyse-it.com/

Great package, and at $175 a year for my Standard Edition requirements
very affordable today. It was a bit less a few years ago. :(

 I can see it as an "always" rule for a consultant who knows that their report benefits from "eye candy" to bullshit an unsophisticated client.

That would be a fatal mistake. NEVER assume the client is not
sophisticated or will not have your work checked.

-hh

--- Synchronet 3.21a-Linux NewsLink 1.2